-
Notifications
You must be signed in to change notification settings - Fork 2
Expand file tree
/
Copy pathdocs.html
More file actions
1046 lines (899 loc) · 69.6 KB
/
docs.html
File metadata and controls
1046 lines (899 loc) · 69.6 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Documentation — openrappter</title>
<meta name="description" content="Complete documentation for openrappter v1.9.1: installation, configuration, agents, memory, skills, gateway, and API reference.">
<meta property="og:title" content="Documentation — openrappter">
<meta property="og:description" content="Complete documentation for the open source local-first AI agent framework with TypeScript and Python runtimes.">
<meta property="og:type" content="website">
<meta property="og:url" content="https://kody-w.github.io/openrappter/docs.html">
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:title" content="Documentation — openrappter">
<meta name="twitter:description" content="Complete documentation for openrappter v1.9.1.">
<link rel="preconnect" href="https://fonts.googleapis.com">
<link href="https://fonts.googleapis.com/css2?family=JetBrains+Mono:wght@400;500;600;700&family=Inter:wght@400;500;600;700;800;900&display=swap" rel="stylesheet">
<link rel="stylesheet" href="./styles.css">
</head>
<body>
<!-- ── Navigation ── -->
<nav>
<div class="nav-container">
<a href="./" class="logo">openrappter <span class="logo-badge">v1.9.1</span></a>
<button class="mobile-menu-btn">☰</button>
<div class="nav-links">
<a href="./docs.html">Docs</a>
<a href="./architecture.html">Architecture</a>
<a href="./tutorial.html">Tutorial</a>
<a href="./changelog.html">Changelog</a>
</div>
<div class="nav-cta">
<a href="https://github.com/kody-w/openrappter" class="btn btn-ghost">
<svg width="18" height="18" viewBox="0 0 24 24" fill="currentColor"><path d="M12 0C5.37 0 0 5.37 0 12c0 5.31 3.435 9.795 8.205 11.385.6.105.825-.255.825-.57 0-.285-.015-1.23-.015-2.235-3.015.555-3.795-.735-4.035-1.41-.135-.345-.72-1.41-1.23-1.695-.42-.225-1.02-.78-.015-.795.945-.015 1.62.87 1.845 1.23 1.08 1.815 2.805 1.305 3.495.99.105-.78.42-1.305.765-1.605-2.67-.3-5.46-1.335-5.46-5.925 0-1.305.465-2.385 1.23-3.225-.12-.3-.54-1.53.12-3.18 0 0 1.005-.315 3.3 1.23.96-.27 1.98-.405 3-.405s2.04.135 3 .405c2.295-1.56 3.3-1.23 3.3-1.23.66 1.65.24 2.88.12 3.18.765.84 1.23 1.905 1.23 3.225 0 4.605-2.805 5.625-5.475 5.925.435.375.81 1.095.81 2.22 0 1.605-.015 2.895-.015 3.3 0 .315.225.69.825.57A12.02 12.02 0 0024 12c0-6.63-5.37-12-12-12z"/></svg>
GitHub
</a>
</div>
</div>
</nav>
<!-- ── Docs Layout ── -->
<div class="docs-layout">
<!-- Sidebar -->
<aside class="sidebar">
<div class="sidebar-title">Getting Started</div>
<ul class="sidebar-nav">
<li><a href="#getting-started">Installation</a></li>
<li><a href="#configuration">Configuration</a></li>
</ul>
<div class="sidebar-title">Core Concepts</div>
<ul class="sidebar-nav">
<li><a href="#agents">Agents Reference</a></li>
<li><a href="#multi-agent">Multi-Agent Patterns</a></li>
<li><a href="#providers">LLM Providers</a></li>
<li><a href="#channels">Messaging Channels</a></li>
<li><a href="#gateway">WebSocket Gateway</a></li>
</ul>
<div class="sidebar-title">Systems</div>
<ul class="sidebar-nav">
<li><a href="#skills">Skills System</a></li>
<li><a href="#memory">Memory System</a></li>
<li><a href="#plugins">Plugin System</a></li>
<li><a href="#security">Security</a></li>
<li><a href="#config-system">Config System</a></li>
</ul>
<div class="sidebar-title">Reference</div>
<ul class="sidebar-nav">
<li><a href="#api">API Reference</a></li>
</ul>
</aside>
<!-- Main Content -->
<main class="docs-content">
<!-- ── 1. Getting Started ── -->
<div id="getting-started" class="doc-section">
<h2>Installation</h2>
<p>OpenRappter is a local-first AI agent framework with parallel implementations in TypeScript and Python. Everything runs on your machine — no cloud account required to get started.</p>
<h3>Prerequisites</h3>
<ul>
<li><strong>Node.js 20+</strong> — required for the TypeScript runtime and CLI</li>
<li><strong>Python 3.10+</strong> — optional, required only if using the Python runtime</li>
<li><strong>GitHub Copilot</strong> — the default zero-config LLM provider (uses your existing subscription)</li>
<li><strong>Git</strong> — required for the clone-based install method</li>
</ul>
<h3>Method 1 — One-line Install (recommended)</h3>
<p>The fastest way to get started. The install script detects your platform, downloads the latest release, and adds <code>openrappter</code> to your PATH.</p>
<pre>curl -fsSL https://kody-w.github.io/openrappter/install.sh | bash</pre>
<p>After the script completes, open a new terminal and verify the install:</p>
<pre>openrappter --version</pre>
<h3>Method 2 — Git Clone</h3>
<p>Clone the repository directly for full access to source code, examples, and the ability to contribute.</p>
<pre>git clone https://github.com/kody-w/openrappter.git
cd openrappter/typescript
npm install
npm run build
npm link # adds openrappter to PATH globally</pre>
<p>For the Python runtime:</p>
<pre>cd openrappter/python
pip install -e .</pre>
<h3>Method 3 — Teach-Your-Agent Install</h3>
<p>If you are already running an AI assistant, you can paste the <a href="https://raw.githubusercontent.com/kody-w/openrappter/main/skills.md">skills.md</a> link into your conversation and ask it to install OpenRappter for you. The agent will follow the instructions in that file, run the install script, and confirm the result — no terminal required on your part.</p>
<h3>Verify</h3>
<pre>openrappter --version
# openrappter v1.9.1
openrappter status
# Runtime: TypeScript/Node 20.x
# Provider: copilot (authenticated)
# Memory: ~/.openrappter/memory.json
# Skills: 0 installed</pre>
<div class="info-box">
<strong>First run:</strong> On first launch openrappter will attempt to authenticate with GitHub Copilot using your existing CLI credentials. If you are not signed in, it will open <code>gh auth login</code> inline without requiring you to run any commands manually.
</div>
</div>
<!-- ── 2. Configuration ── -->
<div id="configuration" class="doc-section">
<h2>Configuration</h2>
<p>OpenRappter reads its configuration from <code>~/.openrappter/config.yaml</code>. The file is created with sensible defaults on first run. All values can be overridden with environment variables.</p>
<h3>Example config.yaml</h3>
<pre># ~/.openrappter/config.yaml
provider:
default: copilot # copilot | anthropic | openai | gemini | ollama
copilot:
model: gpt-4o # model passed to the Copilot API
anthropic:
api_key: ${ANTHROPIC_API_KEY}
model: claude-opus-4-6
openai:
api_key: ${OPENAI_API_KEY}
model: gpt-4o
ollama:
base_url: http://localhost:11434
model: llama3.2
memory:
backend: sqlite # sqlite | json
path: ~/.openrappter/memory.db
embedding_model: all-MiniLM-L6-v2
chunk_size: 512
chunk_overlap: 64
gateway:
enabled: true
port: 8765
host: 127.0.0.1
rate_limit: 100 # requests per minute per connection
shell:
allowlist: [] # if non-empty, only these commands are permitted
blocklist:
- "rm -rf /"
- "sudo rm"
require_approval: false # set true to approve every shell command
skills:
auto_load: true
directory: ~/.openrappter/skills
logging:
level: info # debug | info | warn | error
file: ~/.openrappter/logs/openrappter.log</pre>
<h3>Environment Variable Expansion</h3>
<p>Any value in the YAML can reference environment variables using <code>${VAR_NAME}</code> syntax. The config loader expands these at parse time. This is the recommended way to handle API keys — keep them out of the config file and source them from your shell profile or a <code>.env</code> file.</p>
<h3>Zod Validation</h3>
<p>The configuration schema is defined and validated with Zod v4 at <code>typescript/src/config/schema.ts</code>. On startup, if any required field is missing or has the wrong type, openrappter prints a structured error with the exact path and expected type — not a raw crash.</p>
<h3>Live Reload</h3>
<p>The config system uses a file watcher. When <code>config.yaml</code> is saved, the running process picks up the new values without a restart. Provider keys, log levels, rate limits, and gateway settings all hot-reload. Changes to the memory backend require a restart.</p>
<h3>Environment Variables Reference</h3>
<table class="doc-table">
<thead>
<tr><th>Variable</th><th>Purpose</th><th>Default</th></tr>
</thead>
<tbody>
<tr><td><code>ANTHROPIC_API_KEY</code></td><td>Anthropic Claude API key</td><td>—</td></tr>
<tr><td><code>OPENAI_API_KEY</code></td><td>OpenAI API key</td><td>—</td></tr>
<tr><td><code>GEMINI_API_KEY</code></td><td>Google Gemini API key</td><td>—</td></tr>
<tr><td><code>OPENRAPPTER_CONFIG</code></td><td>Override config file path</td><td><code>~/.openrappter/config.yaml</code></td></tr>
<tr><td><code>OPENRAPPTER_LOG_LEVEL</code></td><td>Override log level</td><td><code>info</code></td></tr>
<tr><td><code>OPENRAPPTER_PORT</code></td><td>Override gateway port</td><td><code>8765</code></td></tr>
<tr><td><code>OPENRAPPTER_PROVIDER</code></td><td>Override default provider</td><td><code>copilot</code></td></tr>
</tbody>
</table>
</div>
<!-- ── 3. Agents Reference ── -->
<div id="agents" class="doc-section">
<h2>Agents Reference</h2>
<p>An agent is a single-responsibility unit of AI-assisted computation. Every agent in OpenRappter is a single file — the metadata contract, documentation, and implementation all live together. There is no YAML, no config file, no magic parsing. The code is the contract.</p>
<h3>Single-File Agent Pattern</h3>
<p>All agents extend <code>BasicAgent</code> and implement one method: <code>perform()</code>. The constructor declares a <code>metadata</code> object that describes the agent's name, purpose, and accepted parameters as a JSON Schema fragment. This metadata is used by the orchestration layer to route requests and validate inputs.</p>
<div class="code-tabs">
<div class="code-tabs-header">
<button class="code-tab-btn active" onclick="switchTab(this,'ts-agent')">TypeScript</button>
<button class="code-tab-btn" onclick="switchTab(this,'py-agent')">Python</button>
</div>
<div id="ts-agent" class="code-tab-content active">
<pre><span class="kw">import</span> { BasicAgent, AgentMetadata } <span class="kw">from</span> <span class="str">'openrappter'</span>;
<span class="kw">export class</span> <span class="typ">MyAgent</span> <span class="kw">extends</span> <span class="typ">BasicAgent</span> {
<span class="fn">constructor</span>() {
<span class="kw">const</span> metadata<span class="op">:</span> <span class="typ">AgentMetadata</span> <span class="op">=</span> {
name<span class="op">:</span> <span class="str">'MyAgent'</span>,
description<span class="op">:</span> <span class="str">'Does something useful'</span>,
parameters<span class="op">:</span> {
type<span class="op">:</span> <span class="str">'object'</span>,
properties<span class="op">:</span> { query<span class="op">:</span> { type<span class="op">:</span> <span class="str">'string'</span> } },
required<span class="op">:</span> [<span class="str">'query'</span>]
}
};
<span class="kw">super</span>(<span class="str">'MyAgent'</span>, metadata);
}
<span class="kw">async</span> <span class="fn">perform</span>(kwargs<span class="op">:</span> Record<span class="op"><</span>string, unknown<span class="op">></span>) {
<span class="kw">const</span> query <span class="op">=</span> kwargs.query <span class="kw">as</span> string;
<span class="cm">// access sloshed context signals</span>
<span class="kw">const</span> timeOfDay <span class="op">=</span> <span class="kw">this</span>.<span class="fn">getSignal</span>(<span class="str">'temporal.time_of_day'</span>);
<span class="kw">return</span> { result<span class="op">:</span> <span class="str">`Hello from MyAgent at ${timeOfDay}`</span> };
}
}</pre>
</div>
<div id="py-agent" class="code-tab-content">
<pre><span class="kw">from</span> openrappter.agents.basic_agent <span class="kw">import</span> BasicAgent
<span class="kw">class</span> <span class="typ">MyAgent</span>(BasicAgent):
<span class="kw">def</span> <span class="fn">__init__</span>(self):
self.name <span class="op">=</span> <span class="str">'MyAgent'</span>
self.metadata <span class="op">=</span> {
<span class="str">"name"</span><span class="op">:</span> self.name,
<span class="str">"description"</span><span class="op">:</span> <span class="str">"Does something useful"</span>,
<span class="str">"parameters"</span><span class="op">:</span> {
<span class="str">"type"</span><span class="op">:</span> <span class="str">"object"</span>,
<span class="str">"properties"</span><span class="op">:</span> {<span class="str">"query"</span><span class="op">:</span> {<span class="str">"type"</span><span class="op">:</span> <span class="str">"string"</span>}},
<span class="str">"required"</span><span class="op">:</span> [<span class="str">"query"</span>]
}
}
<span class="kw">super</span>().__init__(name<span class="op">=</span>self.name, metadata<span class="op">=</span>self.metadata)
<span class="kw">def</span> <span class="fn">perform</span>(self, <span class="op">**</span>kwargs):
query <span class="op">=</span> kwargs.get(<span class="str">"query"</span>, <span class="str">""</span>)
<span class="cm"># access sloshed context signals</span>
time_of_day <span class="op">=</span> self.<span class="fn">get_signal</span>(<span class="str">"temporal.time_of_day"</span>)
<span class="kw">return</span> {<span class="str">"result"</span><span class="op">:</span> <span class="str">f"Hello from MyAgent at {time_of_day}"</span>}</pre>
</div>
</div>
<h3>Execution Flow</h3>
<p>When you call <code>execute(kwargs)</code>, the framework runs this pipeline:</p>
<ol>
<li><code>execute(kwargs)</code> is the public entry point</li>
<li><code>slosh(query)</code> gathers implicit context (temporal signals, query signals, memory echoes, behavioral hints, priors) and synthesizes an <code>Orientation</code> object with confidence score, suggested approach, and contextual hints</li>
<li>Any <code>upstream_slush</code> passed from a previous agent is merged into <code>this.context</code></li>
<li><code>perform(kwargs)</code> is called — this is the method you implement</li>
<li>If the result JSON contains a <code>data_slush</code> key, it is extracted to <code>lastDataSlush</code> (TypeScript) / <code>last_data_slush</code> (Python) for downstream agent chaining</li>
</ol>
<h3>Built-in Agents</h3>
<h4>BasicAgent</h4>
<p>Abstract base class. All agents extend this. Provides data sloshing, signal access, upstream context merging, and the execution pipeline.</p>
<table class="doc-table">
<thead><tr><th>Method</th><th>Description</th></tr></thead>
<tbody>
<tr><td><code>execute(kwargs)</code></td><td>Public entry point. Runs slosh, merges upstream, calls perform.</td></tr>
<tr><td><code>perform(kwargs)</code></td><td>Abstract. Implement this in your subclass.</td></tr>
<tr><td><code>slosh(query)</code></td><td>Gathers implicit context. Called automatically by execute().</td></tr>
<tr><td><code>getSignal(key)</code></td><td>Dot-notation access to sloshed context. e.g. <code>getSignal('temporal.time_of_day')</code></td></tr>
</tbody>
</table>
<h4>ShellAgent</h4>
<p>Executes shell commands and performs file system operations. Supports natural language query parsing so agents upstream can pass a free-text request and ShellAgent will derive the right action.</p>
<table class="doc-table">
<thead><tr><th>Parameter</th><th>Type</th><th>Description</th></tr></thead>
<tbody>
<tr><td><code>action</code></td><td><code>string</code></td><td>One of: <code>bash</code>, <code>read</code>, <code>write</code>, <code>list</code></td></tr>
<tr><td><code>command</code></td><td><code>string</code></td><td>Shell command to run (action: bash)</td></tr>
<tr><td><code>path</code></td><td><code>string</code></td><td>File or directory path (actions: read, write, list)</td></tr>
<tr><td><code>content</code></td><td><code>string</code></td><td>Content to write (action: write)</td></tr>
<tr><td><code>query</code></td><td><code>string</code></td><td>Natural language request — agent will determine action</td></tr>
</tbody>
</table>
<pre><span class="cm">// Execute a shell command</span>
<span class="kw">const</span> result <span class="op">=</span> <span class="kw">await</span> shellAgent.<span class="fn">execute</span>({ action<span class="op">:</span> <span class="str">'bash'</span>, command<span class="op">:</span> <span class="str">'ls -la ~/projects'</span> });
<span class="cm">// Read a file</span>
<span class="kw">const</span> file <span class="op">=</span> <span class="kw">await</span> shellAgent.<span class="fn">execute</span>({ action<span class="op">:</span> <span class="str">'read'</span>, path<span class="op">:</span> <span class="str">'/etc/hosts'</span> });
<span class="cm">// Write a file</span>
<span class="kw">await</span> shellAgent.<span class="fn">execute</span>({ action<span class="op">:</span> <span class="str">'write'</span>, path<span class="op">:</span> <span class="str">'./notes.txt'</span>, content<span class="op">:</span> <span class="str">'hello world'</span> });</pre>
<h4>MemoryAgent</h4>
<p>Stores and retrieves information from the persistent memory store. In TypeScript the backend is SQLite with hybrid search; in Python it is a JSON file at <code>~/.openrappter/memory.json</code>.</p>
<table class="doc-table">
<thead><tr><th>Parameter</th><th>Type</th><th>Description</th></tr></thead>
<tbody>
<tr><td><code>action</code></td><td><code>string</code></td><td>One of: <code>store</code>, <code>recall</code>, <code>forget</code></td></tr>
<tr><td><code>key</code></td><td><code>string</code></td><td>Named identifier for the memory entry</td></tr>
<tr><td><code>content</code></td><td><code>string</code></td><td>Text to store (action: store)</td></tr>
<tr><td><code>query</code></td><td><code>string</code></td><td>Search query for semantic recall (action: recall)</td></tr>
</tbody>
</table>
<h4>Assistant</h4>
<p>A configurable AI assistant with a custom system prompt. Wraps the active LLM provider, forwards sloshed context as additional system context, and streams responses through the gateway when available. Accepts <code>system_prompt</code>, <code>message</code>, and optional <code>conversation_id</code> parameters.</p>
<h4>BrowserAgent</h4>
<p>Headless browser automation powered by Playwright. Supports navigation, clicking, form filling, screenshot capture, and structured data extraction from web pages. Requires <code>playwright</code> to be installed separately (<code>npm install playwright</code> then <code>npx playwright install chromium</code>).</p>
<h4>WebAgent</h4>
<p>HTTP request agent for fetching web content. Supports GET and POST with configurable headers, automatic HTML-to-text extraction, JSON response parsing, and rate-limit-aware retry logic. Does not require a browser runtime — uses native <code>fetch</code> under the hood.</p>
<h4>MessageAgent</h4>
<p>Multi-channel message dispatch. Sends a message to one or more configured channels (Slack, Discord, Telegram, etc.) by channel name. Handles serialization, authentication, and delivery confirmation. Parameters: <code>channel</code>, <code>message</code>, optional <code>thread_id</code>.</p>
<h4>TTSAgent</h4>
<p>Text-to-speech synthesis via <code>edge-tts</code>. Converts text to an MP3 audio file or plays it directly through the system audio output. Supports all Microsoft Edge TTS voices. Parameters: <code>text</code>, <code>voice</code> (e.g. <code>en-US-JennyNeural</code>), optional <code>output_path</code>.</p>
<h4>SessionsAgent</h4>
<p>Session state management for multi-turn conversations. Stores and retrieves keyed session data so agents can maintain context across separate <code>execute()</code> calls without polluting the shared memory store. Parameters: <code>action</code> (<code>get</code>, <code>set</code>, <code>clear</code>), <code>session_id</code>, <code>key</code>, <code>value</code>.</p>
<h4>CronAgent</h4>
<p>Scheduled task execution. Accepts a cron expression and an agent invocation payload. Registers the task with the internal scheduler and fires it at the specified interval. Tasks persist across restarts when the memory backend is SQLite. Parameters: <code>schedule</code> (cron string), <code>agent</code>, <code>kwargs</code>, optional <code>name</code>.</p>
<h4>ImageAgent</h4>
<p>Image processing powered by Sharp. Supports resize, crop, format conversion (JPEG, PNG, WebP, AVIF), metadata extraction, thumbnail generation, and basic filter application. Parameters: <code>action</code>, <code>input_path</code>, <code>output_path</code>, and action-specific options like <code>width</code>, <code>height</code>, <code>format</code>.</p>
<h4>HackerNewsAgent</h4>
<p>Hacker News feed aggregation and summarization. Fetches top stories, new stories, or Ask HN / Show HN posts via the official HN API. Can return raw story data or request an LLM-powered summary of the top items for a digest workflow. Parameters: <code>feed</code> (<code>top</code>, <code>new</code>, <code>ask</code>, <code>show</code>), <code>limit</code>, optional <code>summarize</code> boolean.</p>
<h4>OuroborosAgent</h4>
<p>Self-evolution capability scoring with RPG lineage tracking. Evaluates an agent's output quality across multiple capability dimensions — word statistics, sentiment detection, Caesar cipher encoding, pattern recognition, and reflection accuracy — and produces a quality score from 0–100. Maintains a persistent lineage file tracking scores over time with streak multipliers, evolution tiers, and XP gain. Used for benchmarking model improvements and driving the self-improving agent feedback loop.</p>
<pre><span class="cm">// Evaluate capability quality</span>
<span class="kw">const</span> result <span class="op">=</span> <span class="kw">await</span> ouroborosAgent.<span class="fn">execute</span>({
input<span class="op">:</span> <span class="str">'The quick brown fox jumps over the lazy dog'</span>,
capabilities<span class="op">:</span> [<span class="str">'word_stats'</span>, <span class="str">'sentiment'</span>, <span class="str">'patterns'</span>]
});
<span class="cm">// { score: 87, tier: 'Adept', xp_gained: 43, streak_multiplier: 1.5 }</span></pre>
</div>
<!-- ── 4. Multi-Agent Patterns ── -->
<div id="multi-agent" class="doc-section">
<h2>Multi-Agent Patterns</h2>
<p>OpenRappter provides three composable primitives for coordinating multiple agents: <strong>BroadcastManager</strong> for fan-out execution, <strong>AgentRouter</strong> for rule-based message routing, and <strong>SubAgentManager</strong> for nested hierarchical invocation. All three are available in the TypeScript runtime at <code>typescript/src/agents/</code>.</p>
<h3>BroadcastManager</h3>
<p>Send the same request to multiple agents simultaneously. Three dispatch modes control how results are collected:</p>
<ul>
<li><strong>all</strong> — dispatch to all agents and wait for every response before returning. Results are an array in dispatch order.</li>
<li><strong>race</strong> — dispatch to all agents and return as soon as the first one succeeds. Useful for redundancy or speed-sensitive paths.</li>
<li><strong>fallback</strong> — try agents in order, moving to the next only if the current one fails. Useful for graceful degradation across providers.</li>
</ul>
<pre><span class="kw">import</span> { BroadcastManager } <span class="kw">from</span> <span class="str">'openrappter'</span>;
<span class="kw">const</span> broadcast <span class="op">=</span> <span class="kw">new</span> <span class="typ">BroadcastManager</span>([agentA, agentB, agentC]);
<span class="cm">// Wait for all three</span>
<span class="kw">const</span> allResults <span class="op">=</span> <span class="kw">await</span> broadcast.<span class="fn">send</span>({ query<span class="op">:</span> <span class="str">'status check'</span> }, { mode<span class="op">:</span> <span class="str">'all'</span> });
<span class="cm">// Return the fastest</span>
<span class="kw">const</span> firstResult <span class="op">=</span> <span class="kw">await</span> broadcast.<span class="fn">send</span>({ query<span class="op">:</span> <span class="str">'translate hello'</span> }, { mode<span class="op">:</span> <span class="str">'race'</span> });
<span class="cm">// Try agentA, fall back to agentB, then agentC</span>
<span class="kw">const</span> safeResult <span class="op">=</span> <span class="kw">await</span> broadcast.<span class="fn">send</span>({ query<span class="op">:</span> <span class="str">'generate report'</span> }, { mode<span class="op">:</span> <span class="str">'fallback'</span> });</pre>
<h3>AgentRouter</h3>
<p>Rule-based message routing. Each rule specifies a match condition (sender, channel, group, or regex pattern) and a target agent. Rules are evaluated in priority order. Session key isolation ensures that concurrent conversations on the same router do not cross-contaminate context.</p>
<pre><span class="kw">import</span> { AgentRouter } <span class="kw">from</span> <span class="str">'openrappter'</span>;
<span class="kw">const</span> router <span class="op">=</span> <span class="kw">new</span> <span class="typ">AgentRouter</span>([
{
match<span class="op">:</span> { channel<span class="op">:</span> <span class="str">'slack'</span>, pattern<span class="op">:</span> <span class="str">/^!deploy/</span> },
agent<span class="op">:</span> deployAgent,
priority<span class="op">:</span> <span class="num">10</span>
},
{
match<span class="op">:</span> { sender<span class="op">:</span> <span class="str">'cron'</span> },
agent<span class="op">:</span> schedulerAgent,
priority<span class="op">:</span> <span class="num">5</span>
},
{
match<span class="op">:</span> { group<span class="op">:</span> <span class="str">'default'</span> },
agent<span class="op">:</span> assistantAgent,
priority<span class="op">:</span> <span class="num">0</span>
}
]);
<span class="cm">// Routes to deployAgent (highest priority match)</span>
<span class="kw">await</span> router.<span class="fn">route</span>({
sender<span class="op">:</span> <span class="str">'alice'</span>,
channel<span class="op">:</span> <span class="str">'slack'</span>,
message<span class="op">:</span> <span class="str">'!deploy production'</span>,
session_key<span class="op">:</span> <span class="str">'alice:slack'</span>
});</pre>
<h3>SubAgentManager</h3>
<p>Nested agent invocation with depth limits and loop detection. Allows an agent to spawn child agents as part of its own execution. The manager tracks the call stack and refuses to execute if the same agent appears more than once in the current chain, preventing infinite recursion. The default max depth is 5.</p>
<pre><span class="kw">import</span> { SubAgentManager } <span class="kw">from</span> <span class="str">'openrappter'</span>;
<span class="kw">class</span> <span class="typ">PlannerAgent</span> <span class="kw">extends</span> <span class="typ">BasicAgent</span> {
<span class="kw">private</span> sub<span class="op">:</span> <span class="typ">SubAgentManager</span>;
<span class="fn">constructor</span>() {
<span class="kw">super</span>(<span class="str">'PlannerAgent'</span>, metadata);
<span class="kw">this</span>.sub <span class="op">=</span> <span class="kw">new</span> <span class="typ">SubAgentManager</span>({ maxDepth<span class="op">:</span> <span class="num">3</span> });
}
<span class="kw">async</span> <span class="fn">perform</span>(kwargs) {
<span class="cm">// Invoke a child agent — depth and loop tracking automatic</span>
<span class="kw">const</span> research <span class="op">=</span> <span class="kw">await</span> <span class="kw">this</span>.sub.<span class="fn">invoke</span>(webAgent, {
query<span class="op">:</span> kwargs.topic,
upstream_slush<span class="op">:</span> <span class="kw">this</span>.lastDataSlush
});
<span class="kw">return</span> { plan<span class="op">:</span> research.summary };
}
}</pre>
</div>
<!-- ── 5. LLM Providers ── -->
<div id="providers" class="doc-section">
<h2>LLM Providers</h2>
<p>OpenRappter uses a provider registry to abstract LLM backends. Switching providers is a single config change — agent code never imports provider-specific SDKs directly. The <code>BasicAgent</code> base class exposes a <code>callLLM(messages, options)</code> method that routes to whichever provider is active.</p>
<h3>GitHub Copilot (default)</h3>
<p>Zero-configuration. Uses your existing GitHub Copilot subscription via the <code>gh</code> CLI token. No API key setup required. This is the recommended provider for getting started because authentication is handled inline on first use.</p>
<pre><span class="cm"># config.yaml</span>
provider:
default: copilot
copilot:
model: gpt-4o <span class="cm"># or gpt-4o-mini, claude-3.5-sonnet, o1-mini</span></pre>
<h3>Anthropic</h3>
<p>Access Claude models including claude-opus-4-6, claude-sonnet-4-5, and claude-haiku-3-5. Requires an Anthropic API key.</p>
<pre><span class="cm"># config.yaml</span>
provider:
default: anthropic
anthropic:
api_key: ${ANTHROPIC_API_KEY}
model: claude-opus-4-6
max_tokens: 8192</pre>
<h3>OpenAI</h3>
<p>Supports all GPT-4o and o-series models. Compatible with any OpenAI-compatible endpoint by setting a custom <code>base_url</code>.</p>
<pre><span class="cm"># config.yaml</span>
provider:
default: openai
openai:
api_key: ${OPENAI_API_KEY}
model: gpt-4o
base_url: https://api.openai.com/v1 <span class="cm"># override for compatible endpoints</span></pre>
<h3>Google Gemini</h3>
<p>Access Gemini 2.0 Flash, Gemini 1.5 Pro, and experimental models via Google AI Studio credentials.</p>
<pre><span class="cm"># config.yaml</span>
provider:
default: gemini
gemini:
api_key: ${GEMINI_API_KEY}
model: gemini-2.0-flash-exp</pre>
<h3>Ollama (local)</h3>
<p>Run models entirely locally with no API key or internet connection required. Requires a running Ollama instance. Pull models with <code>ollama pull llama3.2</code> before use.</p>
<pre><span class="cm"># config.yaml</span>
provider:
default: ollama
ollama:
base_url: http://localhost:11434
model: llama3.2 <span class="cm"># any model available in your Ollama instance</span>
timeout: 120000 <span class="cm"># ms — local models can be slow on first load</span></pre>
<h3>Provider Registry Pattern</h3>
<p>Providers are registered at <code>typescript/src/providers/registry.ts</code>. To add a custom provider, implement the <code>LLMProvider</code> interface and register it before starting the agent runtime:</p>
<pre><span class="kw">import</span> { registerProvider } <span class="kw">from</span> <span class="str">'openrappter'</span>;
registerProvider(<span class="str">'my-provider'</span>, {
<span class="kw">async</span> <span class="fn">chat</span>(messages, options) {
<span class="cm">// call your API here</span>
<span class="kw">return</span> { content<span class="op">:</span> response.text };
}
});</pre>
<div class="info-box">
<strong>Tip:</strong> You can configure multiple providers simultaneously and select between them per-agent by passing <code>{ provider: 'anthropic' }</code> to <code>callLLM()</code>. This is useful for routing cheap queries to a fast model and expensive queries to a more capable one.
</div>
</div>
<!-- ── 6. Messaging Channels ── -->
<div id="channels" class="doc-section">
<h2>Messaging Channels</h2>
<p>OpenRappter connects to 15+ messaging platforms through a unified channel interface. All channels implement the same <code>Channel</code> interface: <code>send(message)</code>, <code>receive(handler)</code>, and <code>connect()</code>. Add a channel in <code>config.yaml</code> and it is automatically registered with the router.</p>
<h3>Supported Channels</h3>
<table class="doc-table">
<thead>
<tr><th>Channel</th><th>Auth Method</th><th>Status</th></tr>
</thead>
<tbody>
<tr><td>CLI</td><td>None (local)</td><td>Built-in</td></tr>
<tr><td>Slack</td><td>Bot token + App manifest</td><td>Stable</td></tr>
<tr><td>Discord</td><td>Bot token</td><td>Stable</td></tr>
<tr><td>Telegram</td><td>BotFather token</td><td>Stable</td></tr>
<tr><td>WhatsApp</td><td>Meta Cloud API</td><td>Stable</td></tr>
<tr><td>Signal</td><td>signal-cli daemon</td><td>Beta</td></tr>
<tr><td>Microsoft Teams</td><td>Azure App registration</td><td>Stable</td></tr>
<tr><td>Google Chat</td><td>Service account</td><td>Stable</td></tr>
<tr><td>Matrix</td><td>Access token</td><td>Beta</td></tr>
<tr><td>Mattermost</td><td>Bot token</td><td>Stable</td></tr>
<tr><td>Feishu / Lark</td><td>App credentials</td><td>Beta</td></tr>
<tr><td>Line</td><td>Channel access token</td><td>Beta</td></tr>
<tr><td>Twitch</td><td>OAuth token</td><td>Beta</td></tr>
<tr><td>Nostr</td><td>Private key (nsec)</td><td>Experimental</td></tr>
<tr><td>iMessage</td><td>macOS + AppleScript</td><td>macOS only</td></tr>
</tbody>
</table>
<h3>Channel Configuration</h3>
<pre><span class="cm"># config.yaml</span>
channels:
slack:
enabled: true
bot_token: ${SLACK_BOT_TOKEN}
signing_secret: ${SLACK_SIGNING_SECRET}
default_channel: <span class="str">"#general"</span>
discord:
enabled: true
bot_token: ${DISCORD_BOT_TOKEN}
guild_id: ${DISCORD_GUILD_ID}
telegram:
enabled: true
bot_token: ${TELEGRAM_BOT_TOKEN}
allowed_chat_ids: [] <span class="cm"># empty = allow all chats</span></pre>
<h3>Channel Registry Pattern</h3>
<p>Channels are loaded from <code>typescript/src/channels/</code>. To implement a custom channel, extend the <code>BaseChannel</code> class and register it by name. The channel will then be available in the router and reachable via <code>MessageAgent</code>.</p>
<pre><span class="kw">import</span> { BaseChannel, registerChannel } <span class="kw">from</span> <span class="str">'openrappter'</span>;
<span class="kw">class</span> <span class="typ">MyChannel</span> <span class="kw">extends</span> <span class="typ">BaseChannel</span> {
<span class="kw">async</span> <span class="fn">connect</span>() { <span class="cm">/* establish connection */</span> }
<span class="kw">async</span> <span class="fn">send</span>(message) { <span class="cm">/* deliver message */</span> }
<span class="kw">async</span> <span class="fn">receive</span>(handler) { <span class="cm">/* subscribe to incoming messages */</span> }
}
registerChannel(<span class="str">'my-channel'</span>, MyChannel);</pre>
</div>
<!-- ── 7. WebSocket Gateway ── -->
<div id="gateway" class="doc-section">
<h2>WebSocket Gateway</h2>
<p>The WebSocket gateway provides a real-time bidirectional interface to the agent runtime using the JSON-RPC 2.0 protocol. It enables web frontends, external services, and other processes to invoke agents, subscribe to events, and stream responses without polling.</p>
<h3>Starting the Gateway</h3>
<pre>openrappter gateway start --port 8765</pre>
<p>Or enable it in <code>config.yaml</code> under <code>gateway.enabled: true</code> to start it automatically with the main process.</p>
<h3>Connection Lifecycle</h3>
<ol>
<li>Client connects to <code>ws://localhost:8765</code></li>
<li>Server sends a <code>connected</code> event with session ID and server version</li>
<li>Client sends JSON-RPC requests; server responds with results or streaming chunks</li>
<li>Client subscribes to event channels using <code>subscribe</code> method</li>
<li>Server pushes events to subscribed clients as JSON-RPC notifications</li>
</ol>
<h3>JSON-RPC Request / Response</h3>
<pre><span class="cm">// Request: invoke an agent</span>
{
<span class="str">"jsonrpc"</span><span class="op">:</span> <span class="str">"2.0"</span>,
<span class="str">"id"</span><span class="op">:</span> <span class="str">"req-001"</span>,
<span class="str">"method"</span><span class="op">:</span> <span class="str">"agent.execute"</span>,
<span class="str">"params"</span><span class="op">:</span> {
<span class="str">"agent"</span><span class="op">:</span> <span class="str">"ShellAgent"</span>,
<span class="str">"kwargs"</span><span class="op">:</span> { <span class="str">"action"</span><span class="op">:</span> <span class="str">"bash"</span>, <span class="str">"command"</span><span class="op">:</span> <span class="str">"uptime"</span> },
<span class="str">"stream"</span><span class="op">:</span> <span class="kw">false</span>
}
}
<span class="cm">// Response</span>
{
<span class="str">"jsonrpc"</span><span class="op">:</span> <span class="str">"2.0"</span>,
<span class="str">"id"</span><span class="op">:</span> <span class="str">"req-001"</span>,
<span class="str">"result"</span><span class="op">:</span> {
<span class="str">"output"</span><span class="op">:</span> <span class="str">"12:34 up 3 days, 4:21, 2 users, load averages: 1.23 0.98 0.84"</span>,
<span class="str">"exit_code"</span><span class="op">:</span> <span class="num">0</span>
}
}</pre>
<h3>Streaming Responses</h3>
<p>Set <code>"stream": true</code> in the request params to receive incremental token delivery. The server sends multiple <code>agent.chunk</code> notifications followed by a final <code>agent.done</code> notification with the complete result.</p>
<pre><span class="cm">// Streaming chunk notification</span>
{
<span class="str">"jsonrpc"</span><span class="op">:</span> <span class="str">"2.0"</span>,
<span class="str">"method"</span><span class="op">:</span> <span class="str">"agent.chunk"</span>,
<span class="str">"params"</span><span class="op">:</span> { <span class="str">"id"</span><span class="op">:</span> <span class="str">"req-001"</span>, <span class="str">"delta"</span><span class="op">:</span> <span class="str">"Here is the "</span> }
}
{
<span class="str">"jsonrpc"</span><span class="op">:</span> <span class="str">"2.0"</span>,
<span class="str">"method"</span><span class="op">:</span> <span class="str">"agent.done"</span>,
<span class="str">"params"</span><span class="op">:</span> { <span class="str">"id"</span><span class="op">:</span> <span class="str">"req-001"</span>, <span class="str">"result"</span><span class="op">:</span> { <span class="str">"output"</span><span class="op">:</span> <span class="str">"Here is the analysis..."</span> } }
}</pre>
<h3>Event Types</h3>
<table class="doc-table">
<thead>
<tr><th>Event</th><th>Description</th><th>Subscribe Method</th></tr>
</thead>
<tbody>
<tr><td><code>agent.*</code></td><td>Agent execution lifecycle (start, chunk, done, error)</td><td><code>subscribe.agent</code></td></tr>
<tr><td><code>chat.*</code></td><td>Incoming and outgoing chat messages</td><td><code>subscribe.chat</code></td></tr>
<tr><td><code>channel.*</code></td><td>Channel connect/disconnect/error events</td><td><code>subscribe.channel</code></td></tr>
<tr><td><code>cron.*</code></td><td>Scheduled job fire and completion events</td><td><code>subscribe.cron</code></td></tr>
<tr><td><code>presence.*</code></td><td>Connected client join/leave events</td><td><code>subscribe.presence</code></td></tr>
</tbody>
</table>
<h3>Rate Limiting</h3>
<p>The gateway enforces a per-connection rate limit (default: 100 requests/minute). Exceeding the limit results in a JSON-RPC error response with code <code>-32029</code> (rate limit exceeded) and a <code>retry_after</code> field in milliseconds. Configure the limit in <code>config.yaml</code> under <code>gateway.rate_limit</code>.</p>
</div>
<!-- ── 8. Skills System ── -->
<div id="skills" class="doc-section">
<h2>Skills System</h2>
<p>Skills extend an agent's capabilities at runtime without modifying the agent's source code. A skill is a <code>SKILL.md</code> file that describes a capability in natural language, plus an optional <code>scripts/</code> directory with executable code. Skills are distributed through ClawHub and stored locally at <code>~/.openrappter/skills/</code>.</p>
<h3>ClawHub Integration</h3>
<p>ClawHub is the skills registry. It is accessed via <code>npx clawhub@latest</code> — no global install required. The <code>ClawHubClient</code> in OpenRappter wraps these commands and exposes them as a programmatic API.</p>
<pre><span class="cm"># Search for skills</span>
openrappter skills search <span class="str">"github pull requests"</span>
<span class="cm"># Install a skill by name</span>
openrappter skills install gh-pr-reviewer
<span class="cm"># List installed skills</span>
openrappter skills list
<span class="cm"># Remove a skill</span>
openrappter skills remove gh-pr-reviewer</pre>
<h3>SKILL.md Format</h3>
<p>A skill is defined by a <code>SKILL.md</code> file with a structured frontmatter block followed by documentation:</p>
<pre><span class="str">---
name: gh-pr-reviewer
version: 1.2.0
description: Review GitHub pull requests and suggest improvements
author: kody-w
tags: [github, code-review, productivity]
parameters:
pr_url:
type: string
required: true
description: Full GitHub PR URL to review
focus:
type: string
enum: [security, performance, style, all]
default: all
scripts:
- fetch_pr.sh
- analyze_diff.py
---</span>
<span class="cm"># gh-pr-reviewer
Fetches a pull request diff from GitHub and runs a multi-pass review
focusing on the specified concern area. Returns structured feedback
grouped by file with line-level comments.</span></pre>
<h3>Skill Execution</h3>
<p>When a skill is installed, OpenRappter wraps it as a <code>ClawHubSkillAgent</code> instance that extends <code>BasicAgent</code>. The skill's parameters become the agent's metadata parameters. Scripts in the <code>scripts/</code> directory are executed via <code>ShellAgent</code> with the skill's directory as the working directory.</p>
<pre><span class="kw">import</span> { loadSkill } <span class="kw">from</span> <span class="str">'openrappter'</span>;
<span class="cm">// Skills auto-load at startup if config.skills.auto_load is true</span>
<span class="cm">// Or load manually:</span>
<span class="kw">const</span> prReviewer <span class="op">=</span> <span class="kw">await</span> <span class="fn">loadSkill</span>(<span class="str">'gh-pr-reviewer'</span>);
<span class="kw">const</span> review <span class="op">=</span> <span class="kw">await</span> prReviewer.<span class="fn">execute</span>({
pr_url<span class="op">:</span> <span class="str">'https://github.com/org/repo/pull/42'</span>,
focus<span class="op">:</span> <span class="str">'security'</span>
});</pre>
<h3>Lock File</h3>
<p>Installed skills are tracked in a lock file at <code>~/.openrappter/skills/.clawhub/lock.json</code>. The lock file records the installed version, install date, and a content hash. Running <code>openrappter skills verify</code> checks all installed skills against their hashes and reports any that have been modified locally.</p>
<h3>Writing a Skill</h3>
<p>To publish a skill to ClawHub, create a directory with a <code>SKILL.md</code> file and optional scripts, then run:</p>
<pre>npx clawhub@latest publish ./my-skill-directory</pre>
</div>
<!-- ── 9. Memory System ── -->
<div id="memory" class="doc-section">
<h2>Memory System</h2>
<p>The memory system provides persistent, searchable storage for agent knowledge. It supports both exact key-value retrieval and semantic similarity search via embeddings. The TypeScript runtime uses SQLite with hybrid search; the Python runtime uses a JSON file.</p>
<h3>Content Chunking</h3>
<p>Before storing, large documents are split into overlapping chunks using a sliding window algorithm. This improves recall precision — a query matches the specific passage that contains the answer rather than the entire document. Default settings: 512 token chunks with 64 token overlap.</p>
<pre><span class="cm">// Chunking behavior (typescript/src/memory/chunker.ts)</span>
<span class="cm">// Input: "The quick brown fox..." (2000 tokens)</span>
<span class="cm">// Output: chunks of 512 tokens each, overlapping by 64 tokens</span>
<span class="cm">// Chunk boundaries are respected at sentence boundaries where possible</span>
<span class="kw">const</span> chunker <span class="op">=</span> <span class="kw">new</span> <span class="typ">ContentChunker</span>({ chunkSize<span class="op">:</span> <span class="num">512</span>, overlap<span class="op">:</span> <span class="num">64</span> });
<span class="kw">const</span> chunks <span class="op">=</span> chunker.<span class="fn">split</span>(longDocument);</pre>
<h3>Embeddings</h3>
<p>The TypeScript runtime generates embeddings using a local model (default: <code>all-MiniLM-L6-v2</code> via <code>@xenova/transformers</code>). Embeddings run entirely on-device with no API calls. The first run downloads the model (~25MB) and caches it at <code>~/.openrappter/models/</code>.</p>
<p>To use an API-based embedding provider instead, set <code>memory.embedding_provider: openai</code> in <code>config.yaml</code>. This uses the <code>text-embedding-3-small</code> model by default.</p>
<h3>Hybrid Search</h3>
<p>Memory queries use hybrid search: a combination of vector similarity (cosine distance on embeddings) and keyword BM25 scoring. The two scores are combined with a configurable weight (default: 60% semantic, 40% keyword). This outperforms pure vector search on queries that contain specific identifiers, names, or technical terms.</p>
<pre><span class="cm">// Recall with hybrid search (typescript)</span>
<span class="kw">const</span> results <span class="op">=</span> <span class="kw">await</span> memoryAgent.<span class="fn">execute</span>({
action<span class="op">:</span> <span class="str">'recall'</span>,
query<span class="op">:</span> <span class="str">'deployment configuration for staging'</span>,
limit<span class="op">:</span> <span class="num">5</span>,
min_score<span class="op">:</span> <span class="num">0.7</span>
});</pre>
<h3>Python Backend</h3>
<p>The Python runtime stores all memory entries in a single JSON file at <code>~/.openrappter/memory.json</code>. Recall is performed with a simple TF-IDF + cosine similarity implementation that requires no external dependencies. For high-volume use cases, switch to the SQLite backend by setting <code>memory.backend: sqlite</code> in your config.</p>
<pre><span class="cm"># Python: store and recall</span>
memory_agent.<span class="fn">perform</span>(action<span class="op">=</span><span class="str">"store"</span>, key<span class="op">=</span><span class="str">"project_notes"</span>, content<span class="op">=</span><span class="str">"Deploy to prod on Fridays only."</span>)
results <span class="op">=</span> memory_agent.<span class="fn">perform</span>(action<span class="op">=</span><span class="str">"recall"</span>, query<span class="op">=</span><span class="str">"deployment schedule"</span>)</pre>
<h3>ContextMemoryAgent and ManageMemoryAgent (Python)</h3>
<p>The Python runtime includes two higher-level memory agents built on top of <code>MemoryAgent</code>:</p>
<ul>
<li><strong>ContextMemoryAgent</strong> — Automatically stores conversation turns and retrieves relevant context before each agent call. Designed to be wired into the <code>upstream_slush</code> pipeline.</li>
<li><strong>ManageMemoryAgent</strong> — Exposes CRUD operations for memory management via natural language commands. Useful for admin tasks like bulk deletion, key renaming, and export.</li>
</ul>
</div>
<!-- ── 10. Plugin System ── -->
<div id="plugins" class="doc-section">
<h2>Plugin System</h2>
<p>Plugins extend the OpenRappter runtime itself — adding new channels, providers, storage backends, middleware, or gateway event handlers — without forking the codebase. Plugins are loaded at startup from the <code>~/.openrappter/plugins/</code> directory.</p>
<h3>Plugin Manifest</h3>
<p>Every plugin requires a <code>plugin.json</code> manifest at its root:</p>
<pre>{
<span class="str">"name"</span><span class="op">:</span> <span class="str">"my-plugin"</span>,
<span class="str">"version"</span><span class="op">:</span> <span class="str">"1.0.0"</span>,
<span class="str">"description"</span><span class="op">:</span> <span class="str">"Adds Notion as a memory backend"</span>,
<span class="str">"author"</span><span class="op">:</span> <span class="str">"yourname"</span>,
<span class="str">"main"</span><span class="op">:</span> <span class="str">"./index.js"</span>,
<span class="str">"hooks"</span><span class="op">:</span> [<span class="str">"onLoad"</span>, <span class="str">"onRequest"</span>, <span class="str">"onResponse"</span>],
<span class="str">"permissions"</span><span class="op">:</span> [<span class="str">"memory.write"</span>, <span class="str">"config.read"</span>]
}</pre>
<h3>Plugin Lifecycle</h3>
<p>Plugins implement hooks that are called at defined points in the runtime lifecycle:</p>
<table class="doc-table">
<thead>
<tr><th>Hook</th><th>Called When</th><th>Can Modify</th></tr>
</thead>
<tbody>
<tr><td><code>onLoad</code></td><td>Plugin is first loaded at startup</td><td>Registration, config</td></tr>
<tr><td><code>onUnload</code></td><td>Runtime is shutting down</td><td>Cleanup only</td></tr>
<tr><td><code>onRequest</code></td><td>Before any agent execute() call</td><td>kwargs, context</td></tr>
<tr><td><code>onResponse</code></td><td>After any agent execute() returns</td><td>Result object</td></tr>
<tr><td><code>onMessage</code></td><td>Any channel receives a message</td><td>Message object</td></tr>
<tr><td><code>onError</code></td><td>Any unhandled error in the runtime</td><td>Error handling</td></tr>
</tbody>
</table>
<h3>Example Plugin Structure</h3>
<pre>~/.openrappter/plugins/notion-memory/
plugin.json
index.js
README.md</pre>
<pre><span class="cm">// index.js — minimal plugin implementing onLoad hook</span>
<span class="kw">import</span> { registerStorageAdapter } <span class="kw">from</span> <span class="str">'openrappter/plugins'</span>;
<span class="kw">import</span> { NotionStorageAdapter } <span class="kw">from</span> <span class="str">'./notion-adapter.js'</span>;
<span class="kw">export async function</span> <span class="fn">onLoad</span>(ctx) {
<span class="fn">registerStorageAdapter</span>(<span class="str">'notion'</span>, <span class="kw">new</span> <span class="typ">NotionStorageAdapter</span>({
token<span class="op">:</span> ctx.config.<span class="fn">get</span>(<span class="str">'notion.token'</span>),
databaseId<span class="op">:</span> ctx.config.<span class="fn">get</span>(<span class="str">'notion.database_id'</span>)
}));
ctx.logger.<span class="fn">info</span>(<span class="str">'Notion memory backend registered'</span>);
}</pre>
<h3>SDK Hooks API</h3>
<p>The plugin SDK (<code>openrappter/plugins</code>) exports registration functions for each extensible subsystem: <code>registerStorageAdapter</code>, <code>registerChannel</code>, <code>registerProvider</code>, <code>registerMiddleware</code>, and <code>registerEventHandler</code>. The <code>ctx</code> object passed to each hook provides access to config, logger, memory, and the agent registry.</p>
</div>
<!-- ── 11. Security ── -->
<div id="security" class="doc-section">
<h2>Security</h2>
<p>OpenRappter operates with the same OS-level permissions as the user who runs it. Because agents can execute shell commands and read/write files, the framework includes several layers of protection. Security is opt-in by default for ease of development — enable stricter controls before deploying in shared or production environments.</p>
<h3>ShellAgent Sandboxing</h3>
<p>ShellAgent is the primary attack surface for prompt injection via shell execution. Two mechanisms limit its blast radius:</p>
<ul>
<li><strong>Allowlist mode</strong> — Set <code>shell.allowlist</code> to an array of permitted command prefixes. Any command not matching an entry is rejected before execution. Recommended for production deployments.</li>
<li><strong>Blocklist mode</strong> — The default. A configurable list of dangerous patterns (e.g. <code>rm -rf /</code>, <code>sudo rm</code>) is checked against the command string before execution. Overlapping glob patterns are supported.</li>
</ul>
<pre><span class="cm"># config.yaml — production lockdown example</span>
shell:
allowlist:
- <span class="str">"git status"</span>
- <span class="str">"git log"</span>
- <span class="str">"npm test"</span>
- <span class="str">"ls"</span>
- <span class="str">"cat"</span>
require_approval: false</pre>
<h3>Approval Workflows</h3>
<p>Enable <code>shell.require_approval: true</code> to pause before executing any shell command and prompt the operator for confirmation. In gateway mode this sends an <code>approval.required</code> event to subscribed clients before proceeding. Useful for high-stakes operations like database migrations or deployments.</p>
<h3>Rate Limiting</h3>
<p>The WebSocket gateway enforces per-connection rate limits. For channel adapters, each channel can set its own rate limit in addition to the global gateway limit. Rate limits apply to both inbound message handling and outbound LLM API calls, protecting against runaway loops.</p>
<h3>Audit Logging</h3>
<p>When <code>logging.level: debug</code> or when <code>logging.audit: true</code> is set, all agent invocations, shell commands executed, files written, and channel messages sent are written to the audit log at <code>~/.openrappter/logs/audit.jsonl</code>. Each entry is a newline-delimited JSON object with timestamp, agent name, action, arguments, result status, and duration.</p>
<h3>Gateway Authentication</h3>
<p>By default the gateway binds to <code>127.0.0.1</code> and is not accessible from external networks. For remote access, set a shared secret in <code>config.yaml</code>:</p>
<pre><span class="cm"># config.yaml</span>
gateway:
host: 0.0.0.0
secret: ${GATEWAY_SECRET} <span class="cm"># required in Authorization: Bearer header</span></pre>
<div class="warning-box">
<strong>Warning:</strong> Never expose the gateway on <code>0.0.0.0</code> without setting a <code>gateway.secret</code>. Without a secret, any process on the network can invoke agents and execute shell commands.
</div>
<h3>Prompt Injection Mitigations</h3>
<p>The slosh pipeline includes a simple injection detector that flags inputs containing common injection patterns (e.g. "ignore previous instructions", "you are now DAN"). Flagged inputs are not blocked by default but are logged at <code>warn</code> level and included in the sloshed orientation as a <code>suspicious_input: true</code> signal that agents can act on.</p>
</div>
<!-- ── 12. Config System ── -->
<div id="config-system" class="doc-section">
<h2>Config System</h2>
<p>The config system is responsible for loading, validating, hot-reloading, and providing typed access to all runtime configuration. It is implemented in TypeScript at <code>typescript/src/config/</code> and mirrors the Python implementation at <code>python/openrappter/config.py</code>.</p>
<h3>Loading Order</h3>
<p>Configuration is resolved in this priority order (highest wins):</p>
<ol>
<li>Environment variables (e.g. <code>OPENRAPPTER_LOG_LEVEL=debug</code>)</li>
<li>Config file specified by <code>OPENRAPPTER_CONFIG</code> env var</li>
<li><code>~/.openrappter/config.yaml</code> (default location)</li>
<li><code>~/.openrappter/config.json</code> (JSON alternative)</li>
<li>Built-in defaults compiled into the binary</li>
</ol>
<h3>Zod Schema Validation</h3>
<p>The full config schema is defined in <code>typescript/src/config/schema.ts</code> using Zod v4. Validation runs on every load and reload. Errors are surfaced as structured messages that identify the exact field path and the expected vs received type:</p>
<pre><span class="kw">import</span> { configSchema } <span class="kw">from</span> <span class="str">'openrappter/config'</span>;
<span class="cm">// Validate a config object programmatically</span>
<span class="kw">const</span> result <span class="op">=</span> configSchema.<span class="fn">safeParse</span>(rawConfig);
<span class="kw">if</span> (!result.success) {
result.error.issues.<span class="fn">forEach</span>(issue <span class="op">=></span>
console.<span class="fn">error</span>(<span class="str">`[${issue.path.join('.')}] ${issue.message}`</span>)
);
}</pre>
<h3>File Watcher and Live Reload</h3>
<p>The config loader registers a file system watcher on the config file. When a change is detected, the new file is read, validated, and diffed against the current config. Fields that support hot-reload are updated in place; fields that require a restart (e.g. <code>memory.backend</code>) emit a <code>config.restart_required</code> warning without crashing the process.</p>
<pre><span class="kw">import</span> { getConfig, onConfigChange } <span class="kw">from</span> <span class="str">'openrappter/config'</span>;
<span class="fn">onConfigChange</span>((newConfig, diff) <span class="op">=></span> {
console.<span class="fn">log</span>(<span class="str">'Config updated:'</span>, diff);
<span class="cm">// diff contains only the fields that changed</span>
});</pre>
<h3>Migration System</h3>
<p>As OpenRappter evolves, config schemas change. The migration system tracks the config format version in a <code>_version</code> field and automatically upgrades older config files to the current schema. Migrations are defined as transform functions in <code>typescript/src/config/migrations/</code>. The original file is backed up to <code>config.yaml.bak</code> before any migration is applied.</p>
<h3>Environment Variable Expansion</h3>
<p>Values using <code>${VAR_NAME}</code> syntax are expanded at load time. Unset variables resolve to an empty string by default. Use <code>${VAR_NAME:-default_value}</code> syntax to provide a fallback:</p>
<pre><span class="cm"># Uses ANTHROPIC_API_KEY if set, otherwise 'sk-placeholder'</span>
provider:
anthropic:
api_key: ${ANTHROPIC_API_KEY:-sk-placeholder}</pre>
<h3>Programmatic Config Access</h3>
<pre><span class="kw">import</span> { getConfig } <span class="kw">from</span> <span class="str">'openrappter/config'</span>;
<span class="kw">const</span> config <span class="op">=</span> <span class="fn">getConfig</span>();
console.<span class="fn">log</span>(config.provider.default); <span class="cm">// "copilot"</span>
console.<span class="fn">log</span>(config.gateway.port); <span class="cm">// 8765</span></pre>
</div>
<!-- ── 13. API Reference ── -->
<div id="api" class="doc-section">
<h2>API Reference</h2>
<h3>CLI Commands</h3>
<table class="doc-table">
<thead>
<tr><th>Command</th><th>Description</th></tr>
</thead>
<tbody>
<tr><td><code>openrappter</code></td><td>Start the interactive CLI agent</td></tr>
<tr><td><code>openrappter --version</code></td><td>Print version and exit</td></tr>
<tr><td><code>openrappter status</code></td><td>Show runtime, provider, and memory status</td></tr>
<tr><td><code>openrappter gateway start</code></td><td>Start the WebSocket gateway</td></tr>
<tr><td><code>openrappter gateway stop</code></td><td>Stop a running gateway instance</td></tr>
<tr><td><code>openrappter skills search <query></code></td><td>Search ClawHub for skills</td></tr>
<tr><td><code>openrappter skills install <name></code></td><td>Install a skill from ClawHub</td></tr>
<tr><td><code>openrappter skills list</code></td><td>List installed skills</td></tr>
<tr><td><code>openrappter skills remove <name></code></td><td>Uninstall a skill</td></tr>
<tr><td><code>openrappter skills verify</code></td><td>Verify integrity of installed skills</td></tr>
<tr><td><code>openrappter memory recall <query></code></td><td>Search memory from the CLI</td></tr>
<tr><td><code>openrappter memory clear</code></td><td>Delete all memory entries</td></tr>
<tr><td><code>openrappter config validate</code></td><td>Validate the current config file</td></tr>
</tbody>
</table>
<h3>Programmatic API</h3>
<p>Import OpenRappter as a library in your TypeScript or JavaScript project:</p>
<pre><span class="kw">import</span> {
BasicAgent,
ShellAgent,
MemoryAgent,
BroadcastManager,
AgentRouter,
SubAgentManager,
loadSkill,
getConfig,
onConfigChange,
registerProvider,
registerChannel,
startGateway
} <span class="kw">from</span> <span class="str">'openrappter'</span>;
<span class="cm">// Instantiate and run an agent</span>
<span class="kw">const</span> shell <span class="op">=</span> <span class="kw">new</span> <span class="typ">ShellAgent</span>();
<span class="kw">const</span> result <span class="op">=</span> <span class="kw">await</span> shell.<span class="fn">execute</span>({ action<span class="op">:</span> <span class="str">'bash'</span>, command<span class="op">:</span> <span class="str">'date'</span> });
console.<span class="fn">log</span>(result.output);
<span class="cm">// Start the gateway programmatically</span>
<span class="kw">const</span> gw <span class="op">=</span> <span class="kw">await</span> <span class="fn">startGateway</span>({ port<span class="op">:</span> <span class="num">8765</span> });
gw.<span class="fn">on</span>(<span class="str">'agent.done'</span>, (event) <span class="op">=></span> console.<span class="fn">log</span>(event));</pre>
<h3>Data Sloshing Signals</h3>
<p>The slosh pipeline populates a context object before each <code>perform()</code> call. Access these signals with <code>getSignal(key)</code> using dot notation.</p>
<table class="doc-table">
<thead>
<tr><th>Signal Key</th><th>Type</th><th>Description</th></tr>
</thead>
<tbody>
<tr><td><code>temporal.time_of_day</code></td><td><code>string</code></td><td>morning / afternoon / evening / night</td></tr>
<tr><td><code>temporal.day_of_week</code></td><td><code>string</code></td><td>monday … sunday</td></tr>
<tr><td><code>temporal.iso_date</code></td><td><code>string</code></td><td>ISO 8601 date string</td></tr>
<tr><td><code>temporal.unix_ms</code></td><td><code>number</code></td><td>Current Unix timestamp in milliseconds</td></tr>
<tr><td><code>query.intent</code></td><td><code>string</code></td><td>Inferred intent category of the query</td></tr>
<tr><td><code>query.entities</code></td><td><code>string[]</code></td><td>Named entities extracted from query</td></tr>
<tr><td><code>query.sentiment</code></td><td><code>string</code></td><td>positive / neutral / negative</td></tr>
<tr><td><code>memory.echoes</code></td><td><code>object[]</code></td><td>Top memory recall results for the query</td></tr>
<tr><td><code>memory.recent_keys</code></td><td><code>string[]</code></td><td>Keys written in the last 10 minutes</td></tr>
<tr><td><code>behavioral.session_length</code></td><td><code>number</code></td><td>Number of turns in current session</td></tr>
<tr><td><code>behavioral.last_agent</code></td><td><code>string</code></td><td>Name of last agent called in session</td></tr>
<tr><td><code>orientation.confidence</code></td><td><code>number</code></td><td>0–1 confidence score for chosen approach</td></tr>
<tr><td><code>orientation.approach</code></td><td><code>string</code></td><td>Suggested reasoning approach</td></tr>
<tr><td><code>orientation.hints</code></td><td><code>string[]</code></td><td>Contextual hints derived from slosh</td></tr>
</tbody>
</table>
<h3>Environment Variables</h3>
<table class="doc-table">
<thead>
<tr><th>Variable</th><th>Default</th><th>Description</th></tr>
</thead>
<tbody>
<tr><td><code>ANTHROPIC_API_KEY</code></td><td>—</td><td>Anthropic Claude API key</td></tr>
<tr><td><code>OPENAI_API_KEY</code></td><td>—</td><td>OpenAI API key</td></tr>
<tr><td><code>GEMINI_API_KEY</code></td><td>—</td><td>Google Gemini API key</td></tr>
<tr><td><code>SLACK_BOT_TOKEN</code></td><td>—</td><td>Slack bot token for channel integration</td></tr>
<tr><td><code>DISCORD_BOT_TOKEN</code></td><td>—</td><td>Discord bot token</td></tr>
<tr><td><code>TELEGRAM_BOT_TOKEN</code></td><td>—</td><td>Telegram BotFather token</td></tr>
<tr><td><code>GATEWAY_SECRET</code></td><td>—</td><td>Bearer secret for gateway authentication</td></tr>
<tr><td><code>OPENRAPPTER_CONFIG</code></td><td><code>~/.openrappter/config.yaml</code></td><td>Override config file location</td></tr>
<tr><td><code>OPENRAPPTER_LOG_LEVEL</code></td><td><code>info</code></td><td>debug / info / warn / error</td></tr>
<tr><td><code>OPENRAPPTER_PORT</code></td><td><code>8765</code></td><td>Gateway WebSocket port</td></tr>
<tr><td><code>OPENRAPPTER_PROVIDER</code></td><td><code>copilot</code></td><td>Override default LLM provider</td></tr>
<tr><td><code>OPENRAPPTER_NO_TELEMETRY</code></td><td>—</td><td>Set to any value to disable anonymous usage stats</td></tr>
</tbody>
</table>
<h3>TypeScript Exports</h3>
<pre><span class="cm">// Core agents</span>
<span class="kw">export</span> { BasicAgent, AgentMetadata, AgentResult } <span class="kw">from</span> <span class="str">'./agents/BasicAgent'</span>;