@blockrun/clawrouter 0.12.39 โ 0.12.40
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +129 -115
- package/dist/cli.d.ts +1 -0
- package/dist/cli.js +130 -100
- package/dist/cli.js.map +1 -1
- package/dist/index.d.ts +1387 -0
- package/dist/index.js +127 -97
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -2,17 +2,19 @@
|
|
|
2
2
|
|
|
3
3
|
<img src="assets/banner.png" alt="ClawRouter Banner" width="600">
|
|
4
4
|
|
|
5
|
-
<
|
|
5
|
+
<h1>The LLM router built for autonomous agents</h1>
|
|
6
6
|
|
|
7
|
-
|
|
8
|
-
|
|
9
|
-
|
|
7
|
+
<p>Agents can't sign up for accounts. Agents can't enter credit cards.<br>
|
|
8
|
+
Agents can only sign transactions.<br><br>
|
|
9
|
+
<strong>ClawRouter is the only LLM router that lets agents operate independently.</strong></p>
|
|
10
10
|
|
|
11
|
-
<
|
|
11
|
+
<br>
|
|
12
|
+
|
|
13
|
+
<img src="https://img.shields.io/badge/๐ค_Agent--Native-black?style=for-the-badge" alt="Agent native">
|
|
12
14
|
<img src="https://img.shields.io/badge/๐_Zero_API_Keys-blue?style=for-the-badge" alt="No API keys">
|
|
13
|
-
<img src="https://img.shields.io/badge
|
|
14
|
-
<img src="https://img.shields.io/badge/๐ฐ
|
|
15
|
-
<img src="https://img.shields.io/badge
|
|
15
|
+
<img src="https://img.shields.io/badge/โก_Local_Routing-yellow?style=for-the-badge" alt="Local routing">
|
|
16
|
+
<img src="https://img.shields.io/badge/๐ฐ_x402_USDC-purple?style=for-the-badge" alt="x402 USDC">
|
|
17
|
+
<img src="https://img.shields.io/badge/๐_Open_Source-green?style=for-the-badge" alt="Open source">
|
|
16
18
|
|
|
17
19
|
[](https://npmjs.com/package/@blockrun/clawrouter)
|
|
18
20
|
[](https://npmjs.com/package/@blockrun/clawrouter)
|
|
@@ -32,27 +34,43 @@ One wallet, 41+ models, zero API keys.
|
|
|
32
34
|
|
|
33
35
|
---
|
|
34
36
|
|
|
35
|
-
##
|
|
37
|
+
## Why ClawRouter exists
|
|
38
|
+
|
|
39
|
+
Every other LLM router was built for **human developers** โ create an account, get an API key, pick a model from a dashboard, pay with a credit card.
|
|
40
|
+
|
|
41
|
+
**Agents can't do any of that.**
|
|
42
|
+
|
|
43
|
+
ClawRouter is built for the agent-first world:
|
|
44
|
+
|
|
45
|
+
- **No accounts** โ a wallet is generated locally, no signup
|
|
46
|
+
- **No API keys** โ your wallet signature IS authentication
|
|
47
|
+
- **No model selection** โ 15-dimension scoring picks the right model automatically
|
|
48
|
+
- **No credit cards** โ agents pay per-request with USDC via [x402](https://x402.org)
|
|
49
|
+
- **No trust required** โ runs locally, <1ms routing, zero external dependencies
|
|
50
|
+
|
|
51
|
+
This is the stack that lets agents operate autonomously: **x402 + USDC + local routing**.
|
|
52
|
+
|
|
53
|
+
---
|
|
54
|
+
|
|
55
|
+
## How it compares
|
|
36
56
|
|
|
37
|
-
|
|
|
38
|
-
|
|
|
39
|
-
|
|
|
40
|
-
|
|
|
41
|
-
|
|
|
42
|
-
|
|
|
43
|
-
|
|
|
44
|
-
|
|
|
45
|
-
|
|
|
46
|
-
| [Configuration](#%EF%B8%8F-configuration) | Environment variables |
|
|
47
|
-
| [Troubleshooting](#-troubleshooting) | `doctor` AI-powered diagnostics |
|
|
48
|
-
| [vs OpenRouter](#-vs-openrouter) | Why ClawRouter wins |
|
|
49
|
-
| [Support](#-support) | Telegram, X, founders |
|
|
57
|
+
| | OpenRouter | LiteLLM | Martian | Portkey | **ClawRouter** |
|
|
58
|
+
| ---------------- | ----------------- | ---------------- | ----------------- | ----------------- | ----------------------- |
|
|
59
|
+
| **Models** | 200+ | 100+ | Smart routing | Gateway | **41+** |
|
|
60
|
+
| **Routing** | Manual selection | Manual selection | Smart (closed) | Observability | **Smart (open source)** |
|
|
61
|
+
| **Auth** | Account + API key | Your API keys | Account + API key | Account + API key | **Wallet signature** |
|
|
62
|
+
| **Payment** | Credit card | BYO keys | Credit card | $49-499/mo | **USDC per-request** |
|
|
63
|
+
| **Runs locally** | No | Yes | No | No | **Yes** |
|
|
64
|
+
| **Open source** | No | Yes | No | Partial | **Yes** |
|
|
65
|
+
| **Agent-ready** | No | No | No | No | **Yes** |
|
|
50
66
|
|
|
51
|
-
|
|
67
|
+
โ Open source ยท โ Smart routing ยท โ Runs locally ยท โ Crypto native ยท โ Agent ready
|
|
68
|
+
|
|
69
|
+
**We're the only one that checks all five boxes.**
|
|
52
70
|
|
|
53
71
|
---
|
|
54
72
|
|
|
55
|
-
##
|
|
73
|
+
## Quick Start
|
|
56
74
|
|
|
57
75
|
```bash
|
|
58
76
|
# 1. Install with smart routing enabled
|
|
@@ -63,11 +81,11 @@ openclaw gateway restart
|
|
|
63
81
|
# $5 is enough for thousands of requests
|
|
64
82
|
```
|
|
65
83
|
|
|
66
|
-
Done
|
|
84
|
+
Done. Smart routing (`blockrun/auto`) is now your default model.
|
|
67
85
|
|
|
68
86
|
---
|
|
69
87
|
|
|
70
|
-
##
|
|
88
|
+
## Routing Profiles
|
|
71
89
|
|
|
72
90
|
Choose your routing strategy with `/model <profile>`:
|
|
73
91
|
|
|
@@ -82,7 +100,26 @@ Choose your routing strategy with `/model <profile>`:
|
|
|
82
100
|
|
|
83
101
|
---
|
|
84
102
|
|
|
85
|
-
##
|
|
103
|
+
## How It Works
|
|
104
|
+
|
|
105
|
+
**100% local routing. <1ms latency. Zero external API calls.**
|
|
106
|
+
|
|
107
|
+
```
|
|
108
|
+
Request โ Weighted Scorer (15 dimensions) โ Tier โ Best Model โ Response
|
|
109
|
+
```
|
|
110
|
+
|
|
111
|
+
| Tier | ECO Model | AUTO Model | PREMIUM Model |
|
|
112
|
+
| --------- | ----------------------------------- | ---------------------------- | ---------------------------- |
|
|
113
|
+
| SIMPLE | nvidia/gpt-oss-120b (FREE) | kimi-k2.5 ($0.60/$3.00) | kimi-k2.5 |
|
|
114
|
+
| MEDIUM | gemini-2.5-flash-lite ($0.10/$0.40) | grok-code-fast ($0.20/$1.50) | gpt-5.2-codex ($1.75/$14.00) |
|
|
115
|
+
| COMPLEX | gemini-2.5-flash-lite ($0.10/$0.40) | gemini-3.1-pro ($2/$12) | claude-opus-4.6 ($5/$25) |
|
|
116
|
+
| REASONING | grok-4-fast ($0.20/$0.50) | grok-4-fast ($0.20/$0.50) | claude-sonnet-4.6 ($3/$15) |
|
|
117
|
+
|
|
118
|
+
**Blended average: $2.05/M** vs $25/M for Claude Opus = **92% savings**
|
|
119
|
+
|
|
120
|
+
---
|
|
121
|
+
|
|
122
|
+
## Image Generation
|
|
86
123
|
|
|
87
124
|
Generate images directly from chat with `/imagegen`:
|
|
88
125
|
|
|
@@ -100,16 +137,13 @@ Generate images directly from chat with `/imagegen`:
|
|
|
100
137
|
| `gpt-image` | OpenAI GPT Image 1 | $0.02/image | 1536x1024 |
|
|
101
138
|
| `flux` | Black Forest Flux 1.1 | $0.04/image | 1024x1024 |
|
|
102
139
|
|
|
103
|
-
|
|
104
|
-
|
|
105
|
-
## โ๏ธ Image Editing (img2img)
|
|
140
|
+
## Image Editing (img2img)
|
|
106
141
|
|
|
107
|
-
Edit existing images with `/img2img
|
|
142
|
+
Edit existing images with `/img2img`:
|
|
108
143
|
|
|
109
144
|
```
|
|
110
145
|
/img2img --image ~/photo.png change the background to a starry sky
|
|
111
146
|
/img2img --image ./cat.jpg --mask ./mask.png remove the background
|
|
112
|
-
/img2img --image /tmp/portrait.png --size 1536x1024 add a hat
|
|
113
147
|
```
|
|
114
148
|
|
|
115
149
|
| Option | Required | Description |
|
|
@@ -119,40 +153,11 @@ Edit existing images with `/img2img` โ pass a local file and describe what to
|
|
|
119
153
|
| `--model <model>` | No | Model to use (default: `gpt-image-1`) |
|
|
120
154
|
| `--size <WxH>` | No | Output size (default: `1024x1024`) |
|
|
121
155
|
|
|
122
|
-
|
|
123
|
-
|
|
124
|
-
**API endpoint:** `POST http://localhost:8402/v1/images/image2image` โ accepts local file paths, URLs, or base64 data URIs:
|
|
125
|
-
|
|
126
|
-
```bash
|
|
127
|
-
curl -X POST http://localhost:8402/v1/images/image2image \
|
|
128
|
-
-H "Content-Type: application/json" \
|
|
129
|
-
-d '{"prompt":"add sunglasses","image":"~/photo.png"}'
|
|
130
|
-
```
|
|
131
|
-
|
|
132
|
-
See [Image Generation & Editing docs](docs/image-generation.md#post-v1imagesimage2image) for full API reference and code examples.
|
|
156
|
+
**API endpoint:** `POST http://localhost:8402/v1/images/image2image` โ see [full docs](docs/image-generation.md#post-v1imagesimage2image).
|
|
133
157
|
|
|
134
158
|
---
|
|
135
159
|
|
|
136
|
-
##
|
|
137
|
-
|
|
138
|
-
**100% local routing. <1ms latency. Zero external API calls.**
|
|
139
|
-
|
|
140
|
-
```
|
|
141
|
-
Request โ Weighted Scorer (15 dimensions) โ Tier โ Cheapest Model โ Response
|
|
142
|
-
```
|
|
143
|
-
|
|
144
|
-
| Tier | ECO Model | AUTO Model | PREMIUM Model |
|
|
145
|
-
| --------- | ----------------------------------- | ---------------------------- | ---------------------------- |
|
|
146
|
-
| SIMPLE | nvidia/gpt-oss-120b (FREE) | kimi-k2.5 ($0.60/$3.00) | kimi-k2.5 |
|
|
147
|
-
| MEDIUM | gemini-2.5-flash-lite ($0.10/$0.40) | grok-code-fast ($0.20/$1.50) | gpt-5.2-codex ($1.75/$14.00) |
|
|
148
|
-
| COMPLEX | gemini-2.5-flash-lite ($0.10/$0.40) | gemini-3.1-pro ($2/$12) | claude-opus-4.6 ($5/$25) |
|
|
149
|
-
| REASONING | grok-4-fast ($0.20/$0.50) | grok-4-fast ($0.20/$0.50) | claude-sonnet-4.6 ($3/$15) |
|
|
150
|
-
|
|
151
|
-
**Blended average: $2.05/M** vs $25/M for Claude Opus = **92% savings**
|
|
152
|
-
|
|
153
|
-
---
|
|
154
|
-
|
|
155
|
-
## ๐ฐ Models & Pricing
|
|
160
|
+
## Models & Pricing
|
|
156
161
|
|
|
157
162
|
41+ models across 7 providers, one wallet:
|
|
158
163
|
|
|
@@ -199,31 +204,7 @@ Request โ Weighted Scorer (15 dimensions) โ Tier โ Cheapest Model โ Resp
|
|
|
199
204
|
|
|
200
205
|
---
|
|
201
206
|
|
|
202
|
-
##
|
|
203
|
-
|
|
204
|
-
<table>
|
|
205
|
-
<tr>
|
|
206
|
-
<td width="50%" align="center">
|
|
207
|
-
<strong>Smart Routing in Action</strong><br><br>
|
|
208
|
-
<img src="docs/clawrouter-savings.png" alt="ClawRouter savings" width="400">
|
|
209
|
-
</td>
|
|
210
|
-
<td width="50%" align="center">
|
|
211
|
-
<strong>Telegram Integration</strong><br><br>
|
|
212
|
-
<img src="assets/telegram-demo.png" alt="Telegram demo" width="400">
|
|
213
|
-
</td>
|
|
214
|
-
</tr>
|
|
215
|
-
</table>
|
|
216
|
-
|
|
217
|
-
**The flow:**
|
|
218
|
-
|
|
219
|
-
1. **Wallet auto-generated** on Base (L2) โ saved at `~/.openclaw/blockrun/wallet.key`
|
|
220
|
-
2. **Fund with $1 USDC** โ enough for hundreds of requests
|
|
221
|
-
3. **Request any model** โ ClawRouter picks the cheapest capable one
|
|
222
|
-
4. **Pay per request** โ non-custodial, you hold your keys
|
|
223
|
-
|
|
224
|
-
---
|
|
225
|
-
|
|
226
|
-
## ๐ณ Payment
|
|
207
|
+
## Payment
|
|
227
208
|
|
|
228
209
|
No account. No API key. **Payment IS authentication** via [x402](https://x402.org).
|
|
229
210
|
|
|
@@ -231,9 +212,9 @@ No account. No API key. **Payment IS authentication** via [x402](https://x402.or
|
|
|
231
212
|
Request โ 402 (price: $0.003) โ wallet signs USDC โ retry โ response
|
|
232
213
|
```
|
|
233
214
|
|
|
234
|
-
USDC stays in your wallet until spent
|
|
215
|
+
USDC stays in your wallet until spent โ non-custodial. Price is visible in the 402 header before signing.
|
|
235
216
|
|
|
236
|
-
**Dual-chain support:** Pay with **USDC** on **Base (EVM)** or **USDC on Solana
|
|
217
|
+
**Dual-chain support:** Pay with **USDC** on **Base (EVM)** or **USDC on Solana**. Both wallets are derived from a single BIP-39 mnemonic on first run.
|
|
237
218
|
|
|
238
219
|
```bash
|
|
239
220
|
/wallet # Check balance and address (both chains)
|
|
@@ -251,11 +232,28 @@ USDC stays in your wallet until spent - non-custodial. Price is visible in the 4
|
|
|
251
232
|
- **Base (EVM):** Send USDC on Base to your EVM address
|
|
252
233
|
- **Solana:** Send USDC on Solana to your Solana address
|
|
253
234
|
- **Coinbase/CEX:** Withdraw USDC to either network
|
|
254
|
-
- **Credit card:**
|
|
235
|
+
- **Credit card:** Reach out to [@bc1max on Telegram](https://t.me/bc1max)
|
|
255
236
|
|
|
256
237
|
---
|
|
257
238
|
|
|
258
|
-
##
|
|
239
|
+
## Screenshots
|
|
240
|
+
|
|
241
|
+
<table>
|
|
242
|
+
<tr>
|
|
243
|
+
<td width="50%" align="center">
|
|
244
|
+
<strong>Smart Routing in Action</strong><br><br>
|
|
245
|
+
<img src="docs/clawrouter-savings.png" alt="ClawRouter savings" width="400">
|
|
246
|
+
</td>
|
|
247
|
+
<td width="50%" align="center">
|
|
248
|
+
<strong>Telegram Integration</strong><br><br>
|
|
249
|
+
<img src="assets/telegram-demo.png" alt="Telegram demo" width="400">
|
|
250
|
+
</td>
|
|
251
|
+
</tr>
|
|
252
|
+
</table>
|
|
253
|
+
|
|
254
|
+
---
|
|
255
|
+
|
|
256
|
+
## Configuration
|
|
259
257
|
|
|
260
258
|
For basic usage, no configuration needed. For advanced options:
|
|
261
259
|
|
|
@@ -270,22 +268,7 @@ For basic usage, no configuration needed. For advanced options:
|
|
|
270
268
|
|
|
271
269
|
---
|
|
272
270
|
|
|
273
|
-
##
|
|
274
|
-
|
|
275
|
-
| | OpenRouter / LiteLLM | ClawRouter |
|
|
276
|
-
| --------------- | --------------------------- | -------------------------------- |
|
|
277
|
-
| **Setup** | Human creates account | Agent generates wallet |
|
|
278
|
-
| **Auth** | API key (shared secret) | Wallet signature (cryptographic) |
|
|
279
|
-
| **Payment** | Prepaid balance (custodial) | Per-request (non-custodial) |
|
|
280
|
-
| **Routing** | Proprietary / closed | Open source, client-side |
|
|
281
|
-
| **Rate limits** | Per-key quotas | None (your wallet, your limits) |
|
|
282
|
-
| **Cost** | $25/M (Opus equivalent) | $2.05/M blended average |
|
|
283
|
-
|
|
284
|
-
**[Full comparison โ](docs/vs-openrouter.md)**
|
|
285
|
-
|
|
286
|
-
---
|
|
287
|
-
|
|
288
|
-
## ๐ฉบ Troubleshooting
|
|
271
|
+
## Troubleshooting
|
|
289
272
|
|
|
290
273
|
**When things go wrong, run the doctor:**
|
|
291
274
|
|
|
@@ -333,7 +316,7 @@ npx @blockrun/clawrouter doctor opus "ๆทฑๅบฆๅๆๆ็้
็ฝฎ"
|
|
|
333
316
|
|
|
334
317
|
---
|
|
335
318
|
|
|
336
|
-
##
|
|
319
|
+
## Development
|
|
337
320
|
|
|
338
321
|
```bash
|
|
339
322
|
git clone https://github.com/BlockRunAI/ClawRouter.git
|
|
@@ -345,7 +328,7 @@ npm test
|
|
|
345
328
|
|
|
346
329
|
---
|
|
347
330
|
|
|
348
|
-
##
|
|
331
|
+
## Support
|
|
349
332
|
|
|
350
333
|
| Channel | Link |
|
|
351
334
|
| --------------------- | ------------------------------------------------------------------ |
|
|
@@ -357,7 +340,40 @@ npm test
|
|
|
357
340
|
|
|
358
341
|
---
|
|
359
342
|
|
|
360
|
-
##
|
|
343
|
+
## From the BlockRun Ecosystem
|
|
344
|
+
|
|
345
|
+
<table>
|
|
346
|
+
<tr>
|
|
347
|
+
<td width="50%">
|
|
348
|
+
|
|
349
|
+
### โก ClawRouter
|
|
350
|
+
|
|
351
|
+
**The LLM router built for autonomous agents**
|
|
352
|
+
|
|
353
|
+
You're here. 41+ models, local smart routing, x402 USDC payments โ the only stack that lets agents operate independently.
|
|
354
|
+
|
|
355
|
+
`curl -fsSL https://blockrun.ai/ClawRouter-update | bash`
|
|
356
|
+
|
|
357
|
+
</td>
|
|
358
|
+
<td width="50%">
|
|
359
|
+
|
|
360
|
+
### ๐ฆ [SocialClaw](https://github.com/BlockRunAI/socialclaw)
|
|
361
|
+
|
|
362
|
+
**Intelligence-as-a-function for X/Twitter**
|
|
363
|
+
|
|
364
|
+
The first X analytics an agent can call. One function call = one intelligence report. $0.08, not $49/month. No dashboard, no login, no subscription.
|
|
365
|
+
|
|
366
|
+
`pip install blockrun-llm[solana]`
|
|
367
|
+
|
|
368
|
+
[](https://github.com/BlockRunAI/socialclaw)
|
|
369
|
+
|
|
370
|
+
</td>
|
|
371
|
+
</tr>
|
|
372
|
+
</table>
|
|
373
|
+
|
|
374
|
+
---
|
|
375
|
+
|
|
376
|
+
## More Resources
|
|
361
377
|
|
|
362
378
|
| Resource | Description |
|
|
363
379
|
| ------------------------------------------------------ | ------------------------ |
|
|
@@ -367,16 +383,14 @@ npm test
|
|
|
367
383
|
| [Routing Profiles](docs/routing-profiles.md) | ECO/AUTO/PREMIUM details |
|
|
368
384
|
| [Architecture](docs/architecture.md) | Technical deep dive |
|
|
369
385
|
| [Configuration](docs/configuration.md) | Environment variables |
|
|
370
|
-
| [vs OpenRouter](docs/vs-openrouter.md) | Why ClawRouter wins |
|
|
371
|
-
| [Features](docs/features.md) | All features |
|
|
372
386
|
| [Troubleshooting](docs/troubleshooting.md) | Common issues |
|
|
373
387
|
|
|
374
388
|
---
|
|
375
389
|
|
|
376
390
|
<div align="center">
|
|
377
391
|
|
|
378
|
-
**MIT License** ยท [BlockRun](https://blockrun.ai) โ
|
|
392
|
+
**MIT License** ยท [BlockRun](https://blockrun.ai) โ Agent-native AI infrastructure
|
|
379
393
|
|
|
380
|
-
โญ If ClawRouter
|
|
394
|
+
โญ If ClawRouter powers your agents, consider starring the repo!
|
|
381
395
|
|
|
382
396
|
</div>
|
package/dist/cli.d.ts
ADDED
|
@@ -0,0 +1 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
package/dist/cli.js
CHANGED
|
@@ -525,6 +525,99 @@ function getFallbackChainFiltered(tier, tierConfigs, estimatedTotalTokens, getCo
|
|
|
525
525
|
return filtered;
|
|
526
526
|
}
|
|
527
527
|
|
|
528
|
+
// src/router/strategy.ts
|
|
529
|
+
var RulesStrategy = class {
|
|
530
|
+
name = "rules";
|
|
531
|
+
route(prompt, systemPrompt, maxOutputTokens, options) {
|
|
532
|
+
const { config, modelPricing } = options;
|
|
533
|
+
const fullText = `${systemPrompt ?? ""} ${prompt}`;
|
|
534
|
+
const estimatedTokens = Math.ceil(fullText.length / 4);
|
|
535
|
+
const ruleResult = classifyByRules(prompt, systemPrompt, estimatedTokens, config.scoring);
|
|
536
|
+
const { routingProfile } = options;
|
|
537
|
+
let tierConfigs;
|
|
538
|
+
let profileSuffix;
|
|
539
|
+
let profile;
|
|
540
|
+
if (routingProfile === "eco" && config.ecoTiers) {
|
|
541
|
+
tierConfigs = config.ecoTiers;
|
|
542
|
+
profileSuffix = " | eco";
|
|
543
|
+
profile = "eco";
|
|
544
|
+
} else if (routingProfile === "premium" && config.premiumTiers) {
|
|
545
|
+
tierConfigs = config.premiumTiers;
|
|
546
|
+
profileSuffix = " | premium";
|
|
547
|
+
profile = "premium";
|
|
548
|
+
} else {
|
|
549
|
+
const agenticScore = ruleResult.agenticScore ?? 0;
|
|
550
|
+
const isAutoAgentic = agenticScore >= 0.5;
|
|
551
|
+
const isExplicitAgentic = config.overrides.agenticMode ?? false;
|
|
552
|
+
const hasToolsInRequest = options.hasTools ?? false;
|
|
553
|
+
const useAgenticTiers = (hasToolsInRequest || isAutoAgentic || isExplicitAgentic) && config.agenticTiers != null;
|
|
554
|
+
tierConfigs = useAgenticTiers ? config.agenticTiers : config.tiers;
|
|
555
|
+
profileSuffix = useAgenticTiers ? ` | agentic${hasToolsInRequest ? " (tools)" : ""}` : "";
|
|
556
|
+
profile = useAgenticTiers ? "agentic" : "auto";
|
|
557
|
+
}
|
|
558
|
+
const agenticScoreValue = ruleResult.agenticScore;
|
|
559
|
+
if (estimatedTokens > config.overrides.maxTokensForceComplex) {
|
|
560
|
+
const decision2 = selectModel(
|
|
561
|
+
"COMPLEX",
|
|
562
|
+
0.95,
|
|
563
|
+
"rules",
|
|
564
|
+
`Input exceeds ${config.overrides.maxTokensForceComplex} tokens${profileSuffix}`,
|
|
565
|
+
tierConfigs,
|
|
566
|
+
modelPricing,
|
|
567
|
+
estimatedTokens,
|
|
568
|
+
maxOutputTokens,
|
|
569
|
+
routingProfile,
|
|
570
|
+
agenticScoreValue
|
|
571
|
+
);
|
|
572
|
+
return { ...decision2, tierConfigs, profile };
|
|
573
|
+
}
|
|
574
|
+
const hasStructuredOutput = systemPrompt ? /json|structured|schema/i.test(systemPrompt) : false;
|
|
575
|
+
let tier;
|
|
576
|
+
let confidence;
|
|
577
|
+
const method = "rules";
|
|
578
|
+
let reasoning = `score=${ruleResult.score.toFixed(2)} | ${ruleResult.signals.join(", ")}`;
|
|
579
|
+
if (ruleResult.tier !== null) {
|
|
580
|
+
tier = ruleResult.tier;
|
|
581
|
+
confidence = ruleResult.confidence;
|
|
582
|
+
} else {
|
|
583
|
+
tier = config.overrides.ambiguousDefaultTier;
|
|
584
|
+
confidence = 0.5;
|
|
585
|
+
reasoning += ` | ambiguous -> default: ${tier}`;
|
|
586
|
+
}
|
|
587
|
+
if (hasStructuredOutput) {
|
|
588
|
+
const tierRank = { SIMPLE: 0, MEDIUM: 1, COMPLEX: 2, REASONING: 3 };
|
|
589
|
+
const minTier = config.overrides.structuredOutputMinTier;
|
|
590
|
+
if (tierRank[tier] < tierRank[minTier]) {
|
|
591
|
+
reasoning += ` | upgraded to ${minTier} (structured output)`;
|
|
592
|
+
tier = minTier;
|
|
593
|
+
}
|
|
594
|
+
}
|
|
595
|
+
reasoning += profileSuffix;
|
|
596
|
+
const decision = selectModel(
|
|
597
|
+
tier,
|
|
598
|
+
confidence,
|
|
599
|
+
method,
|
|
600
|
+
reasoning,
|
|
601
|
+
tierConfigs,
|
|
602
|
+
modelPricing,
|
|
603
|
+
estimatedTokens,
|
|
604
|
+
maxOutputTokens,
|
|
605
|
+
routingProfile,
|
|
606
|
+
agenticScoreValue
|
|
607
|
+
);
|
|
608
|
+
return { ...decision, tierConfigs, profile };
|
|
609
|
+
}
|
|
610
|
+
};
|
|
611
|
+
var registry = /* @__PURE__ */ new Map();
|
|
612
|
+
registry.set("rules", new RulesStrategy());
|
|
613
|
+
function getStrategy(name) {
|
|
614
|
+
const strategy = registry.get(name);
|
|
615
|
+
if (!strategy) {
|
|
616
|
+
throw new Error(`Unknown routing strategy: ${name}`);
|
|
617
|
+
}
|
|
618
|
+
return strategy;
|
|
619
|
+
}
|
|
620
|
+
|
|
528
621
|
// src/router/config.ts
|
|
529
622
|
var DEFAULT_ROUTING_CONFIG = {
|
|
530
623
|
version: "2.0",
|
|
@@ -1617,7 +1710,11 @@ var DEFAULT_ROUTING_CONFIG = {
|
|
|
1617
1710
|
SIMPLE: {
|
|
1618
1711
|
primary: "nvidia/gpt-oss-120b",
|
|
1619
1712
|
// FREE! $0.00/$0.00
|
|
1620
|
-
fallback: [
|
|
1713
|
+
fallback: [
|
|
1714
|
+
"google/gemini-2.5-flash-lite",
|
|
1715
|
+
"google/gemini-2.5-flash",
|
|
1716
|
+
"deepseek/deepseek-chat"
|
|
1717
|
+
]
|
|
1621
1718
|
},
|
|
1622
1719
|
MEDIUM: {
|
|
1623
1720
|
primary: "google/gemini-2.5-flash-lite",
|
|
@@ -1742,77 +1839,8 @@ var DEFAULT_ROUTING_CONFIG = {
|
|
|
1742
1839
|
|
|
1743
1840
|
// src/router/index.ts
|
|
1744
1841
|
function route(prompt, systemPrompt, maxOutputTokens, options) {
|
|
1745
|
-
const
|
|
1746
|
-
|
|
1747
|
-
const estimatedTokens = Math.ceil(fullText.length / 4);
|
|
1748
|
-
const ruleResult = classifyByRules(prompt, systemPrompt, estimatedTokens, config.scoring);
|
|
1749
|
-
const { routingProfile } = options;
|
|
1750
|
-
let tierConfigs;
|
|
1751
|
-
let profileSuffix;
|
|
1752
|
-
if (routingProfile === "eco" && config.ecoTiers) {
|
|
1753
|
-
tierConfigs = config.ecoTiers;
|
|
1754
|
-
profileSuffix = " | eco";
|
|
1755
|
-
} else if (routingProfile === "premium" && config.premiumTiers) {
|
|
1756
|
-
tierConfigs = config.premiumTiers;
|
|
1757
|
-
profileSuffix = " | premium";
|
|
1758
|
-
} else {
|
|
1759
|
-
const agenticScore = ruleResult.agenticScore ?? 0;
|
|
1760
|
-
const isAutoAgentic = agenticScore >= 0.5;
|
|
1761
|
-
const isExplicitAgentic = config.overrides.agenticMode ?? false;
|
|
1762
|
-
const hasToolsInRequest = options.hasTools ?? false;
|
|
1763
|
-
const useAgenticTiers = (hasToolsInRequest || isAutoAgentic || isExplicitAgentic) && config.agenticTiers != null;
|
|
1764
|
-
tierConfigs = useAgenticTiers ? config.agenticTiers : config.tiers;
|
|
1765
|
-
profileSuffix = useAgenticTiers ? ` | agentic${hasToolsInRequest ? " (tools)" : ""}` : "";
|
|
1766
|
-
}
|
|
1767
|
-
const agenticScoreValue = ruleResult.agenticScore;
|
|
1768
|
-
if (estimatedTokens > config.overrides.maxTokensForceComplex) {
|
|
1769
|
-
return selectModel(
|
|
1770
|
-
"COMPLEX",
|
|
1771
|
-
0.95,
|
|
1772
|
-
"rules",
|
|
1773
|
-
`Input exceeds ${config.overrides.maxTokensForceComplex} tokens${profileSuffix}`,
|
|
1774
|
-
tierConfigs,
|
|
1775
|
-
modelPricing,
|
|
1776
|
-
estimatedTokens,
|
|
1777
|
-
maxOutputTokens,
|
|
1778
|
-
routingProfile,
|
|
1779
|
-
agenticScoreValue
|
|
1780
|
-
);
|
|
1781
|
-
}
|
|
1782
|
-
const hasStructuredOutput = systemPrompt ? /json|structured|schema/i.test(systemPrompt) : false;
|
|
1783
|
-
let tier;
|
|
1784
|
-
let confidence;
|
|
1785
|
-
const method = "rules";
|
|
1786
|
-
let reasoning = `score=${ruleResult.score.toFixed(2)} | ${ruleResult.signals.join(", ")}`;
|
|
1787
|
-
if (ruleResult.tier !== null) {
|
|
1788
|
-
tier = ruleResult.tier;
|
|
1789
|
-
confidence = ruleResult.confidence;
|
|
1790
|
-
} else {
|
|
1791
|
-
tier = config.overrides.ambiguousDefaultTier;
|
|
1792
|
-
confidence = 0.5;
|
|
1793
|
-
reasoning += ` | ambiguous -> default: ${tier}`;
|
|
1794
|
-
}
|
|
1795
|
-
if (hasStructuredOutput) {
|
|
1796
|
-
const tierRank = { SIMPLE: 0, MEDIUM: 1, COMPLEX: 2, REASONING: 3 };
|
|
1797
|
-
const minTier = config.overrides.structuredOutputMinTier;
|
|
1798
|
-
if (tierRank[tier] < tierRank[minTier]) {
|
|
1799
|
-
reasoning += ` | upgraded to ${minTier} (structured output)`;
|
|
1800
|
-
tier = minTier;
|
|
1801
|
-
}
|
|
1802
|
-
}
|
|
1803
|
-
reasoning += profileSuffix;
|
|
1804
|
-
return selectModel(
|
|
1805
|
-
tier,
|
|
1806
|
-
confidence,
|
|
1807
|
-
method,
|
|
1808
|
-
reasoning,
|
|
1809
|
-
tierConfigs,
|
|
1810
|
-
modelPricing,
|
|
1811
|
-
estimatedTokens,
|
|
1812
|
-
maxOutputTokens,
|
|
1813
|
-
routingProfile,
|
|
1814
|
-
agenticScoreValue
|
|
1815
|
-
);
|
|
1842
|
+
const strategy = getStrategy("rules");
|
|
1843
|
+
return strategy.route(prompt, systemPrompt, maxOutputTokens, options);
|
|
1816
1844
|
}
|
|
1817
1845
|
|
|
1818
1846
|
// src/models.ts
|
|
@@ -5026,6 +5054,12 @@ var ROUTING_PROFILES = /* @__PURE__ */ new Set([
|
|
|
5026
5054
|
"premium"
|
|
5027
5055
|
]);
|
|
5028
5056
|
var FREE_MODEL = "nvidia/gpt-oss-120b";
|
|
5057
|
+
var FREE_TIER_CONFIGS = {
|
|
5058
|
+
SIMPLE: { primary: FREE_MODEL, fallback: [] },
|
|
5059
|
+
MEDIUM: { primary: FREE_MODEL, fallback: [] },
|
|
5060
|
+
COMPLEX: { primary: FREE_MODEL, fallback: [] },
|
|
5061
|
+
REASONING: { primary: FREE_MODEL, fallback: [] }
|
|
5062
|
+
};
|
|
5029
5063
|
var freeRequestCount = 0;
|
|
5030
5064
|
var MAX_MESSAGES = 200;
|
|
5031
5065
|
var CONTEXT_LIMIT_KB = 5120;
|
|
@@ -6900,7 +6934,17 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
|
|
|
6900
6934
|
|
|
6901
6935
|
`;
|
|
6902
6936
|
}
|
|
6903
|
-
routingDecision = {
|
|
6937
|
+
routingDecision = {
|
|
6938
|
+
model: freeModel,
|
|
6939
|
+
tier: "SIMPLE",
|
|
6940
|
+
confidence: 1,
|
|
6941
|
+
method: "rules",
|
|
6942
|
+
reasoning: "free profile",
|
|
6943
|
+
costEstimate: 0,
|
|
6944
|
+
baselineCost: 0,
|
|
6945
|
+
savings: 1,
|
|
6946
|
+
tierConfigs: FREE_TIER_CONFIGS
|
|
6947
|
+
};
|
|
6904
6948
|
} else {
|
|
6905
6949
|
effectiveSessionId = getSessionId(req.headers) ?? deriveSessionId(parsedMessages);
|
|
6906
6950
|
const existingSession = effectiveSessionId ? sessionStore.getSession(effectiveSessionId) : void 0;
|
|
@@ -6991,18 +7035,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
|
|
|
6991
7035
|
const contentHash = hashRequestContent(prompt, toolCallNames);
|
|
6992
7036
|
const shouldEscalate = sessionStore.recordRequestHash(effectiveSessionId, contentHash);
|
|
6993
7037
|
if (shouldEscalate) {
|
|
6994
|
-
const activeTierConfigs =
|
|
6995
|
-
if (routingDecision.reasoning?.includes("agentic") && routerOpts.config.agenticTiers) {
|
|
6996
|
-
return routerOpts.config.agenticTiers;
|
|
6997
|
-
}
|
|
6998
|
-
if (routingProfile === "eco" && routerOpts.config.ecoTiers) {
|
|
6999
|
-
return routerOpts.config.ecoTiers;
|
|
7000
|
-
}
|
|
7001
|
-
if (routingProfile === "premium" && routerOpts.config.premiumTiers) {
|
|
7002
|
-
return routerOpts.config.premiumTiers;
|
|
7003
|
-
}
|
|
7004
|
-
return routerOpts.config.tiers;
|
|
7005
|
-
})();
|
|
7038
|
+
const activeTierConfigs = routingDecision.tierConfigs ?? routerOpts.config.tiers;
|
|
7006
7039
|
const escalation = sessionStore.escalateSession(
|
|
7007
7040
|
effectiveSessionId,
|
|
7008
7041
|
activeTierConfigs
|
|
@@ -7218,18 +7251,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
|
|
|
7218
7251
|
if (routingDecision) {
|
|
7219
7252
|
const estimatedInputTokens = Math.ceil(body.length / 4);
|
|
7220
7253
|
const estimatedTotalTokens = estimatedInputTokens + maxTokens;
|
|
7221
|
-
const tierConfigs =
|
|
7222
|
-
if (routingDecision.reasoning?.includes("agentic") && routerOpts.config.agenticTiers) {
|
|
7223
|
-
return routerOpts.config.agenticTiers;
|
|
7224
|
-
}
|
|
7225
|
-
if (routingProfile === "eco" && routerOpts.config.ecoTiers) {
|
|
7226
|
-
return routerOpts.config.ecoTiers;
|
|
7227
|
-
}
|
|
7228
|
-
if (routingProfile === "premium" && routerOpts.config.premiumTiers) {
|
|
7229
|
-
return routerOpts.config.premiumTiers;
|
|
7230
|
-
}
|
|
7231
|
-
return routerOpts.config.tiers;
|
|
7232
|
-
})();
|
|
7254
|
+
const tierConfigs = routingDecision.tierConfigs ?? routerOpts.config.tiers;
|
|
7233
7255
|
const fullChain = getFallbackChain(routingDecision.tier, tierConfigs);
|
|
7234
7256
|
const contextFiltered = getFallbackChainFiltered(
|
|
7235
7257
|
routingDecision.tier,
|
|
@@ -7309,6 +7331,14 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
|
|
|
7309
7331
|
status: result.errorStatus || 500
|
|
7310
7332
|
};
|
|
7311
7333
|
if (result.isProviderError && !isLastAttempt) {
|
|
7334
|
+
const isExplicitModelError = !routingDecision;
|
|
7335
|
+
const isUnknownExplicitModel = isExplicitModelError && /unknown.*model|invalid.*model/i.test(result.errorBody || "");
|
|
7336
|
+
if (isUnknownExplicitModel) {
|
|
7337
|
+
console.log(
|
|
7338
|
+
`[ClawRouter] Explicit model error from ${tryModel}, not falling back: ${result.errorBody?.slice(0, 100)}`
|
|
7339
|
+
);
|
|
7340
|
+
break;
|
|
7341
|
+
}
|
|
7312
7342
|
if (result.errorStatus === 429) {
|
|
7313
7343
|
markRateLimited(tryModel);
|
|
7314
7344
|
try {
|
|
@@ -8084,8 +8114,8 @@ Commands:
|
|
|
8084
8114
|
partners List available partner APIs with pricing
|
|
8085
8115
|
partners test Test partner API endpoints (expect 402 = alive)
|
|
8086
8116
|
wallet recover Restore wallet.key from mnemonic (if generated by ClawRouter)
|
|
8087
|
-
chain solana Switch
|
|
8088
|
-
chain base Switch
|
|
8117
|
+
chain solana Switch to Solana (persists). Aliases: /wallet solana, wallet solana
|
|
8118
|
+
chain base Switch to Base EVM (persists). Aliases: /wallet base, wallet base
|
|
8089
8119
|
|
|
8090
8120
|
Examples:
|
|
8091
8121
|
# Start standalone proxy
|
|
@@ -8155,7 +8185,7 @@ function parseArgs(args) {
|
|
|
8155
8185
|
} else if (arg === "wallet" && args[i + 1] === "recover") {
|
|
8156
8186
|
result.walletRecover = true;
|
|
8157
8187
|
i++;
|
|
8158
|
-
} else if (arg === "chain" && (args[i + 1] === "solana" || args[i + 1] === "base")) {
|
|
8188
|
+
} else if ((arg === "chain" || arg === "/wallet" || arg === "wallet") && (args[i + 1] === "solana" || args[i + 1] === "base")) {
|
|
8159
8189
|
result.chain = args[i + 1];
|
|
8160
8190
|
i++;
|
|
8161
8191
|
} else if (arg === "--port" && args[i + 1]) {
|