copilot-custom-endpoint 1.4.2 → 1.5.0

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 CHANGED
@@ -21,9 +21,9 @@ That's it. No code, no servers to manage (unless the model specifically needs th
21
21
 
22
22
  | Model | Provider | Needs proxy? | Vision | Setup guide |
23
23
  | --------------------------- | --------- | ---------------------- | ------------ | -------------------------------------------------------------------------------------------------- |
24
- | **MiMo V2 Flash** | Xiaomi | No | ❌ | [Setup](docs/models/mimo.md) |
25
- | **MiMo V2.5** | Xiaomi | No | ✅ | [Setup](docs/models/mimo.md) |
26
- | **MiMo V2.5 Pro** | Xiaomi | No | ❌ | [Setup](docs/models/mimo.md) |
24
+ | **MiMo V2 Flash** | Xiaomi | Optional (recommended) | ❌ | [Setup](docs/models/mimo.md) |
25
+ | **MiMo V2.5** | Xiaomi | Optional (recommended) | ✅ | [Setup](docs/models/mimo.md) |
26
+ | **MiMo V2.5 Pro** | Xiaomi | Optional (recommended) | ❌ | [Setup](docs/models/mimo.md) |
27
27
  | **Kimi K2.7 Code / K2.6** | Moonshot | **Yes** | ✅ | [Setup](docs/models/kimi.md) |
28
28
  | **Qwen 3.7 Plus** | DashScope | Optional (recommended) | ✅ | [Setup](docs/models/qwen.md) |
29
29
  | **Qwen 3.7 Max** | DashScope | Optional (recommended) | ❌ | [Setup](docs/models/qwen.md) |
@@ -70,9 +70,10 @@ If a model needs a proxy, the setup guide will tell you to run a command first.
70
70
  Run from the repo root:
71
71
 
72
72
  ```bash
73
- npm run proxy # Start both proxies (Kimi + Qwen)
73
+ npm run proxy # Start all proxies (Kimi + Qwen + MiMo)
74
74
  npm run proxy:kimi # Start only the Kimi proxy
75
75
  npm run proxy:qwen # Start only the Qwen proxy
76
+ npm run proxy:mimo # Start only the MiMo proxy
76
77
  npm run clean:logs # Remove debug_log/
77
78
  npm test # Run the test suite
78
79
  ```
@@ -80,9 +81,10 @@ npm test # Run the test suite
80
81
  Or globally via npx (no clone needed):
81
82
 
82
83
  ```bash
83
- npx copilot-custom-endpoint # Start both proxies
84
+ npx copilot-custom-endpoint # Start all proxies
84
85
  npx copilot-custom-endpoint kimi # Kimi only
85
86
  npx copilot-custom-endpoint qwen # Qwen only
87
+ npx copilot-custom-endpoint mimo # MiMo only
86
88
  npx copilot-custom-endpoint clean # Remove debug_log/
87
89
  ```
88
90
 
package/cli.mjs CHANGED
@@ -8,13 +8,14 @@ import { rmSync } from 'node:fs'
8
8
  const __dirname = dirname(fileURLToPath(import.meta.url))
9
9
  const sub = process.argv[2]
10
10
 
11
- const usage = `Usage: copilot-custom-endpoint [all|kimi|qwen|clean]
11
+ const usage = `Usage: copilot-custom-endpoint [all|kimi|qwen|mimo|clean]
12
12
 
13
13
  Start a local proxy for VS Code Copilot custom endpoints.
14
14
 
15
- copilot-custom-endpoint all Start both proxies concurrently (default)
15
+ copilot-custom-endpoint all Start all proxies concurrently (default)
16
16
  copilot-custom-endpoint kimi Start the Kimi K2 proxy on port 3457
17
17
  copilot-custom-endpoint qwen Start the Qwen 3.x proxy on port 3458
18
+ copilot-custom-endpoint mimo Start the MiMo V2.5 proxy on port 3459
18
19
  copilot-custom-endpoint clean Remove the debug_log/ directory
19
20
 
20
21
  Environment variables: see --help for each proxy.
@@ -26,12 +27,18 @@ if (sub === 'clean') {
26
27
  process.exit(0)
27
28
  }
28
29
 
29
- if (sub && sub !== 'kimi' && sub !== 'qwen' && sub !== 'all') {
30
+ if (
31
+ sub &&
32
+ sub !== 'kimi' &&
33
+ sub !== 'qwen' &&
34
+ sub !== 'mimo' &&
35
+ sub !== 'all'
36
+ ) {
30
37
  console.error(usage)
31
38
  process.exit(1)
32
39
  }
33
40
 
34
- const targets = sub === 'all' || !sub ? ['kimi', 'qwen'] : [sub]
41
+ const targets = sub === 'all' || !sub ? ['kimi', 'qwen', 'mimo'] : [sub]
35
42
 
36
43
  // Spawn all target proxies and wait for all to exit.
37
44
  // This keeps both proxies alive in "all" mode instead of exiting
@@ -1,10 +1,10 @@
1
1
  # Full example config
2
2
 
3
- Here's a complete, real-world `chatLanguageModels.json` that combines **all the providers documented in this repo**. Copy what you need, leave the rest out.
3
+ Here's a complete, real-world `chatLanguageModels.json` that combines **the `customendpoint` providers from the live `chatLanguageModels.json`**. Copy what you need, leave the rest out.
4
4
 
5
5
  > **Note:** The `apiKey` fields are left as empty strings — set them via the **Chat: Manage Language Models** UI (Command Palette → right-click provider group → **Update API Key**). After you set a key via the UI, VS Code replaces the empty string with a `${input:chat.lm.secret.<id>}` secret reference.
6
6
  >
7
- > This combined config reflects the same provider blocks as the live `chatLanguageModels.json`. Qwen is pointed at the local proxy; remove `requestBody.enable_thinking` when using the proxy.
7
+ > The live config points Qwen at the local proxy (`:3458`) and MiMo at the local proxy (`:3459`). When using a proxy, align the model `requestBody` overrides with the proxy's behavior: Qwen sends no `requestBody` (the proxy manages `enable_thinking` dynamically); MiMo sends only `temperature` and `top_p` (the proxy injects `thinking: {"type": "disabled"}` on tool turns and leaves it absent on plain chat).
8
8
 
9
9
  ```json
10
10
  [
@@ -76,14 +76,13 @@ Here's a complete, real-world `chatLanguageModels.json` that combines **all the
76
76
  {
77
77
  "id": "mimo-v2.5-pro",
78
78
  "name": "MiMo V2.5 Pro (text)",
79
- "url": "https://api.xiaomimimo.com/v1/chat/completions",
79
+ "url": "http://127.0.0.1:3459/v1/chat/completions",
80
80
  "toolCalling": true,
81
81
  "vision": false,
82
82
  "streaming": true,
83
83
  "maxInputTokens": 1048576,
84
84
  "maxOutputTokens": 131072,
85
85
  "requestBody": {
86
- "thinking": { "type": "disabled" },
87
86
  "temperature": 1,
88
87
  "top_p": 0.95
89
88
  }
@@ -91,14 +90,13 @@ Here's a complete, real-world `chatLanguageModels.json` that combines **all the
91
90
  {
92
91
  "id": "mimo-v2.5",
93
92
  "name": "MiMo V2.5 (vision)",
94
- "url": "https://api.xiaomimimo.com/v1/chat/completions",
93
+ "url": "http://127.0.0.1:3459/v1/chat/completions",
95
94
  "toolCalling": true,
96
95
  "vision": true,
97
96
  "streaming": true,
98
97
  "maxInputTokens": 1048576,
99
98
  "maxOutputTokens": 32768,
100
99
  "requestBody": {
101
- "thinking": { "type": "disabled" },
102
100
  "temperature": 1,
103
101
  "top_p": 0.95
104
102
  }
@@ -106,14 +104,13 @@ Here's a complete, real-world `chatLanguageModels.json` that combines **all the
106
104
  {
107
105
  "id": "mimo-v2-flash",
108
106
  "name": "MiMo V2 Flash (text)",
109
- "url": "https://api.xiaomimimo.com/v1/chat/completions",
107
+ "url": "http://127.0.0.1:3459/v1/chat/completions",
110
108
  "toolCalling": true,
111
109
  "vision": false,
112
110
  "streaming": true,
113
111
  "maxInputTokens": 262144,
114
112
  "maxOutputTokens": 65536,
115
113
  "requestBody": {
116
- "thinking": { "type": "disabled" },
117
114
  "temperature": 0.3,
118
115
  "top_p": 0.95
119
116
  }
@@ -195,4 +192,4 @@ If you only need one provider, jump straight to its setup guide:
195
192
  - [MiniMax M3](minimax.md)
196
193
  - [GLM (5.1 / 5V Turbo)](glm.md)
197
194
 
198
- > **DeepSeek V4 Pro / V4 Flash** use the [DeepSeek V4 for Copilot Chat](https://marketplace.visualstudio.com/items?itemName=Vizards.deepseek-v4-for-copilot) extension — they don't appear in `chatLanguageModels.json`.
195
+ > **DeepSeek V4 Pro / V4 Flash** use the [DeepSeek V4 for Copilot Chat](https://marketplace.visualstudio.com/items?itemName=Vizards.deepseek-v4-for-copilot) extension. They appear in `chatLanguageModels.json` as `vendor: "deepseek"` (not as a `customendpoint` provider) and are configured via the extension's settings block.
@@ -76,7 +76,7 @@ Config file location:
76
76
  },
77
77
  {
78
78
  "id": "kimi-k2.7-code",
79
- "name": "Kimi K2.7 Code",
79
+ "name": "Kimi K2.7 Code (vision)",
80
80
  "url": "http://127.0.0.1:3457/v1/chat/completions",
81
81
  "requestBody": {
82
82
  "temperature": 1,
@@ -1,18 +1,18 @@
1
1
  # Xiaomi MiMo — VS Code Custom Endpoint Setup Guide
2
2
 
3
- > **TL;DR:** MiMo works directly — no proxy needed. Set `thinking: { type: "disabled" }` in `requestBody` for tool-loop stability, because MiMo's API rejects (HTTP 400) any tool turn that is missing historical `reasoning_content`. Disabling thinking eliminates the field, so loops stay stable.
3
+ > **TL;DR:** MiMo works direct with static `thinking: { type: "disabled" }` in `requestBody`, or via `proxy/mimo-proxy.mjs` for dynamic thinking suppression (reasoning visible in plain chat, suppressed in tool loops). Static suppression is simpler; the proxy lets you see reasoning in non-agent chats.
4
4
 
5
5
  ## At a Glance
6
6
 
7
- | Field | Value |
8
- | ---------------------- | --------------------------------------------------- |
9
- | Mode | **Direct** (no proxy) |
10
- | Vision | ✅ Yes (`mimo-v2.5` only) |
11
- | Tool calling | ✅ Yes (with `thinking: disabled`) |
12
- | Context | 1M (V2.5 Pro / V2.5) / 256K (V2 Flash) |
13
- | Max output | 131072 (V2.5 Pro) / 32768 (V2.5) / 65536 (V2 Flash) |
14
- | Required `requestBody` | `thinking: { type: "disabled" }` |
15
- | Endpoint | `https://api.xiaomimimo.com/v1/chat/completions` |
7
+ | Field | Value |
8
+ | ---------------------- | -------------------------------------------------------------------------------------------------------------- |
9
+ | Mode | **Direct** (proxy optional) |
10
+ | Vision | ✅ Yes (`mimo-v2.5` only) |
11
+ | Tool calling | ✅ Yes (with `thinking: disabled`) |
12
+ | Context | 1M (V2.5 Pro / V2.5) / 256K (V2 Flash) |
13
+ | Max output | 131072 (V2.5 Pro) / 32768 (V2.5) / 65536 (V2 Flash) |
14
+ | Required `requestBody` | Direct: `thinking: { type: "disabled" }`<br>Proxy: none (proxy handles it) |
15
+ | Endpoint | Direct: `https://api.xiaomimimo.com/v1/chat/completions`<br>Proxy: `http://127.0.0.1:3459/v1/chat/completions` |
16
16
 
17
17
  ### Models at a glance
18
18
 
@@ -26,10 +26,21 @@
26
26
 
27
27
  ## Quick Start
28
28
 
29
+ ### Direct (no proxy)
30
+
29
31
  1. **Edit `chatLanguageModels.json`** — add the MiMo block(s) from [Setup](#setup) below.
30
32
  2. **Set your `MIMO_API_KEY`** via Command Palette → **Chat: Manage Language Models**.
31
33
  3. **Restart VS Code** and pick "MiMo V2.5 Pro", "MiMo V2.5", or "MiMo V2 Flash".
32
34
 
35
+ ### With optional proxy (dynamic thinking)
36
+
37
+ The `proxy/mimo-proxy.mjs` provides dynamic thinking suppression: reasoning stays ON in plain chat but turns OFF automatically when tools are invoked.
38
+
39
+ - `npm run proxy:mimo` (or `node proxy/mimo-proxy.mjs`) starts the local proxy on `http://127.0.0.1:3459/v1/chat/completions`.
40
+ - `node proxy/mimo-proxy.mjs --help` prints the supported environment variables and defaults.
41
+
42
+ When using the proxy, update your model URLs to `http://127.0.0.1:3459/v1/chat/completions` and **remove** `thinking` from `requestBody`. The proxy handles it dynamically.
43
+
33
44
  ## Setup
34
45
 
35
46
  ### 1. VS Code configuration
@@ -42,6 +53,8 @@ Config file location:
42
53
  | macOS | `~/Library/Application Support/Code/User/chatLanguageModels.json` |
43
54
  | Linux | `~/.config/Code/User/chatLanguageModels.json` |
44
55
 
56
+ #### Direct (static `thinking: disabled`)
57
+
45
58
  ```json
46
59
  {
47
60
  "name": "MiMo",
@@ -98,6 +111,61 @@ Config file location:
98
111
  }
99
112
  ```
100
113
 
114
+ #### With proxy (dynamic thinking, no `requestBody` override)
115
+
116
+ ```json
117
+ {
118
+ "name": "MiMo (proxy)",
119
+ "vendor": "customendpoint",
120
+ "apiKey": "",
121
+ "apiType": "chat-completions",
122
+ "models": [
123
+ {
124
+ "id": "mimo-v2.5-pro",
125
+ "name": "MiMo V2.5 Pro (text, proxy)",
126
+ "url": "http://127.0.0.1:3459/v1/chat/completions",
127
+ "toolCalling": true,
128
+ "vision": false,
129
+ "streaming": true,
130
+ "maxInputTokens": 1048576,
131
+ "maxOutputTokens": 131072,
132
+ "requestBody": {
133
+ "temperature": 1,
134
+ "top_p": 0.95
135
+ }
136
+ },
137
+ {
138
+ "id": "mimo-v2.5",
139
+ "name": "MiMo V2.5 (vision, proxy)",
140
+ "url": "http://127.0.0.1:3459/v1/chat/completions",
141
+ "toolCalling": true,
142
+ "vision": true,
143
+ "streaming": true,
144
+ "maxInputTokens": 1048576,
145
+ "maxOutputTokens": 32768,
146
+ "requestBody": {
147
+ "temperature": 1,
148
+ "top_p": 0.95
149
+ }
150
+ },
151
+ {
152
+ "id": "mimo-v2-flash",
153
+ "name": "MiMo V2 Flash (text, proxy)",
154
+ "url": "http://127.0.0.1:3459/v1/chat/completions",
155
+ "toolCalling": true,
156
+ "vision": false,
157
+ "streaming": true,
158
+ "maxInputTokens": 262144,
159
+ "maxOutputTokens": 65536,
160
+ "requestBody": {
161
+ "temperature": 0.3,
162
+ "top_p": 0.95
163
+ }
164
+ }
165
+ ]
166
+ }
167
+ ```
168
+
101
169
  ### 2. API key
102
170
 
103
171
  1. Open the Command Palette (`Ctrl+Shift+P`).
@@ -156,13 +224,13 @@ When thinking is enabled, responses include a `reasoning_content` field alongsid
156
224
 
157
225
  ## Troubleshooting
158
226
 
159
- | Symptom | Likely cause | Fix |
160
- | ------------------------------------------ | -------------------------------------------------------- | ----------------------------------------------------- |
161
- | HTTP 400 on the second turn of a tool loop | `reasoning_content` missing in history (thinking on) | Add `thinking: { type: "disabled" }` to `requestBody` |
162
- | Vision request returns an error | Used `mimo-v2.5-pro` or `mimo-v2-flash` (text-only) | Use `mimo-v2.5` for vision |
163
- | Custom `tool_choice` ignored | MiMo only honors `"auto"` | Stick to `auto` |
164
- | 401 Unauthorized | Wrong key, or Token Plan URL used with pay-as-you-go key | Match key prefix (`sk-` vs `tp-`) to the endpoint |
165
- | 429 rate-limited | Concurrent sessions exceeded 100 RPM / 10M TPM | Reduce concurrent agent sessions |
227
+ | Symptom | Likely cause | Fix |
228
+ | ------------------------------------------ | -------------------------------------------------------- | ---------------------------------------------------------------------------------------------- |
229
+ | HTTP 400 on the second turn of a tool loop | `reasoning_content` missing in history (thinking on) | Add `thinking: { type: "disabled" }` to `requestBody`, or use the proxy (`npm run proxy:mimo`) |
230
+ | Vision request returns an error | Used `mimo-v2.5-pro` or `mimo-v2-flash` (text-only) | Use `mimo-v2.5` for vision |
231
+ | Custom `tool_choice` ignored | MiMo only honors `"auto"` | Stick to `auto` |
232
+ | 401 Unauthorized | Wrong key, or Token Plan URL used with pay-as-you-go key | Match key prefix (`sk-` vs `tp-`) to the endpoint |
233
+ | 429 rate-limited | Concurrent sessions exceeded 100 RPM / 10M TPM | Reduce concurrent agent sessions |
166
234
 
167
235
  ## Pricing
168
236
 
@@ -194,14 +262,14 @@ This is the same class of problem as Qwen's `reasoning_content` issue, but **str
194
262
  - **Thinking disabled + tool calling = works** (no `reasoning_content` to preserve).
195
263
  - **Thinking enabled + plain chat = works** (no tool calls in history).
196
264
 
197
- ### Why a static `thinking: disabled` is enough
265
+ ### Proxy for dynamic thinking suppression
198
266
 
199
- VS Code's agent mode is the only flow that triggers tool loops, and we already disable thinking for those turns. Plain chat with thinking enabled works fine because no `reasoning_content` accumulates in history.
267
+ A dynamic proxy (`proxy/mimo-proxy.mjs`) is now available — same pattern as `proxy/qwen-proxy.mjs`. It suppresses thinking only when tools are present, letting plain chat show reasoning while keeping tool loops stable.
200
268
 
201
- A dynamic proxy (suppress thinking only when tools are present — same pattern as `proxy/qwen-proxy.mjs`) would let plain chat show reasoning, but it is **not implemented** because:
269
+ - **Plain chat** → `thinking` is removed from the request body, so MiMo uses its API default (enabled for V2.5 Pro / V2.5).
270
+ - **Tool-enabled requests** → `thinking: { type: "disabled" }` is injected, preventing `reasoning_content` 400 errors.
202
271
 
203
- - The cost of losing visible reasoning in plain chat is low for most users.
204
- - Static suppression is one less moving part to maintain.
272
+ Static suppression (direct mode) remains a perfectly valid simpler alternative.
205
273
 
206
274
  ### Benchmark highlights (from official MiMo V2.5 announcement)
207
275
 
@@ -233,13 +301,13 @@ External API checks (curl):
233
301
 
234
302
  ### Known risks
235
303
 
236
- | Risk | Detail | Mitigation |
237
- | ------------------------------------- | ------------------------------------------------------------------ | ------------------------------------------------------- |
238
- | `reasoning_content` 400 errors | If thinking is accidentally enabled in tool loops, API returns 400 | Always set `thinking.type: "disabled"` in `requestBody` |
239
- | `tool_choice` only supports `"auto"` | Non-`auto` values are stripped | Should not affect VS Code, which uses `auto` |
240
- | Auth header format | Both `api-key:` and `Authorization: Bearer` work | VS Code sends `Authorization: Bearer` — works directly |
241
- | `temperature` locked in thinking mode | V2.5 Pro / V2.5 force `temperature: 1.0` when thinking is on | Not an issue when thinking is disabled |
242
- | 1M context window | VS Code may not send enough tokens to benefit | Set conservatively; adjust after testing |
304
+ | Risk | Detail | Mitigation |
305
+ | ------------------------------------- | ------------------------------------------------------------------ | -------------------------------------------------------------------------- |
306
+ | `reasoning_content` 400 errors | If thinking is accidentally enabled in tool loops, API returns 400 | Set `thinking.type: "disabled"` in `requestBody` (direct) or use the proxy |
307
+ | `tool_choice` only supports `"auto"` | Non-`auto` values are stripped | Should not affect VS Code, which uses `auto` |
308
+ | Auth header format | Both `api-key:` and `Authorization: Bearer` work | VS Code sends `Authorization: Bearer` — works directly |
309
+ | `temperature` locked in thinking mode | V2.5 Pro / V2.5 force `temperature: 1.0` when thinking is on | Not an issue when thinking is disabled |
310
+ | 1M context window | VS Code may not send enough tokens to benefit | Set conservatively; adjust after testing |
243
311
 
244
312
  ## References
245
313
 
package/package.json CHANGED
@@ -1,13 +1,14 @@
1
1
  {
2
2
  "name": "copilot-custom-endpoint",
3
- "version": "1.4.2",
4
- "description": "Local proxies for VS Code Copilot custom endpoints — Kimi K2 & Qwen 3.x",
3
+ "version": "1.5.0",
4
+ "description": "Local proxies for VS Code Copilot custom endpoints — Kimi K2, Qwen 3.x & MiMo V2.5",
5
5
  "license": "MIT",
6
6
  "type": "module",
7
7
  "bin": {
8
8
  "copilot-custom-endpoint": "cli.mjs",
9
9
  "copilot-custom-endpoint-kimi": "proxy/kimi-proxy.mjs",
10
- "copilot-custom-endpoint-qwen": "proxy/qwen-proxy.mjs"
10
+ "copilot-custom-endpoint-qwen": "proxy/qwen-proxy.mjs",
11
+ "copilot-custom-endpoint-mimo": "proxy/mimo-proxy.mjs"
11
12
  },
12
13
  "files": [
13
14
  "README.md",
@@ -19,9 +20,10 @@
19
20
  "proxy/"
20
21
  ],
21
22
  "scripts": {
22
- "proxy": "concurrently --names kimi,qwen --prefix-colors cyan,green \"npm run proxy:kimi\" \"npm run proxy:qwen\"",
23
+ "proxy": "concurrently --names kimi,qwen,mimo --prefix-colors cyan,green,magenta \"npm run proxy:kimi\" \"npm run proxy:qwen\" \"npm run proxy:mimo\"",
23
24
  "proxy:kimi": "node proxy/kimi-proxy.mjs",
24
25
  "proxy:qwen": "node proxy/qwen-proxy.mjs",
26
+ "proxy:mimo": "node proxy/mimo-proxy.mjs",
25
27
  "clean:logs": "node -e \"import{rmSync}from'node:fs';rmSync('debug_log',{recursive:true,force:true});console.log('debug_log/ removed')\"",
26
28
  "test": "node --test tests/**/*.test.mjs",
27
29
  "lint": "eslint .",
@@ -0,0 +1,127 @@
1
+ #!/usr/bin/env node
2
+ import 'dotenv/config'
3
+ import { fileURLToPath } from 'node:url'
4
+ import { createProxy } from '../lib/create-proxy.mjs'
5
+
6
+ /**
7
+ * Supported model scope for this proxy:
8
+ * - Validated with `mimo-v2.5-pro`, `mimo-v2.5`, and `mimo-v2-flash`.
9
+ * - Expected to work for any MiMo model that supports the `thinking` object
10
+ * with a `type` field on the OpenAI-compatible surface.
11
+ * - Not intended for non-MiMo providers, because the rewrite assumes
12
+ * MiMo's `thinking.type` behavior.
13
+ */
14
+ const upstreamUrl =
15
+ process.env.MIMO_UPSTREAM_URL ??
16
+ 'https://api.xiaomimimo.com/v1/chat/completions'
17
+ const port = Number.parseInt(
18
+ process.env.MIMO_PROXY_PORT ?? process.env.PORT ?? '3459',
19
+ 10
20
+ )
21
+ const disableThinkingWithTools =
22
+ (process.env.MIMO_PROXY_DISABLE_THINKING_WITH_TOOLS ?? '1') !== '0'
23
+ const defaultLogPath = fileURLToPath(
24
+ new URL('../debug_log/mimo-proxy.ndjson', import.meta.url)
25
+ )
26
+ const logPath = process.env.MIMO_PROXY_LOG ?? defaultLogPath
27
+
28
+ if (process.argv.includes('--help')) {
29
+ console.log(`MiMo proxy
30
+
31
+ Starts a local HTTP proxy that conditionally injects thinking: { type: "disabled" }
32
+ when the request includes a tools array, letting MiMo models show reasoning in
33
+ plain chat while keeping tool loops stable.
34
+
35
+ Environment variables:
36
+ MIMO_PROXY_PORT Local listen port. Default: 3459 (falls back to PORT)
37
+ MIMO_UPSTREAM_URL Upstream MiMo chat-completions URL.
38
+ Default: https://api.xiaomimimo.com/v1/chat/completions
39
+ MIMO_PROXY_DISABLE_THINKING_WITH_TOOLS
40
+ Inject thinking: { type: "disabled" } when tools are present.
41
+ Default: 1
42
+ MIMO_PROXY_LOG Path to the redacted NDJSON log file.
43
+
44
+ Suggested VS Code model URL:
45
+ http://127.0.0.1:3459/v1/chat/completions
46
+ `)
47
+ process.exit(0)
48
+ }
49
+
50
+ // ---- Provider-specific rewrite logic ----
51
+
52
+ function summarizePayload(payload, hasTools, rewriteInfo) {
53
+ const messages = Array.isArray(payload.messages) ? payload.messages : []
54
+ const tools = Array.isArray(payload.tools) ? payload.tools : []
55
+
56
+ return {
57
+ model: payload.model,
58
+ stream: payload.stream,
59
+ hasTools,
60
+ toolCount: tools.length,
61
+ toolChoice: payload.tool_choice,
62
+ ...rewriteInfo,
63
+ maxTokens:
64
+ payload.max_tokens ??
65
+ payload.max_completion_tokens ??
66
+ payload.max_output_tokens,
67
+ messageCount: messages.length,
68
+ messageRoles: messages.map((message) => message?.role).slice(0, 16),
69
+ topLevelKeys: Object.keys(payload).sort()
70
+ }
71
+ }
72
+
73
+ function rewriteMiMo(payload) {
74
+ // Determine if a tool is actually being invoked:
75
+ // - tool_choice is set and not "none"
76
+ // - OR there is a "tool" role message in the conversation
77
+ const messages = Array.isArray(payload.messages) ? payload.messages : []
78
+ const hasToolRole = messages.some((message) => message?.role === 'tool')
79
+ const toolChoice = payload.tool_choice
80
+ const hasActiveToolCall =
81
+ hasToolRole ||
82
+ (toolChoice !== undefined && toolChoice !== 'none' && toolChoice !== null)
83
+ const hasTools = hasActiveToolCall
84
+ const incomingThinkingType = payload?.thinking?.type
85
+
86
+ if (disableThinkingWithTools && hasTools) {
87
+ // Tool-enabled request: suppress thinking to avoid reasoning_content issues
88
+ payload.thinking = { type: 'disabled' }
89
+ } else {
90
+ // Plain chat: remove thinking so the model uses its default (enabled for V2.5 models)
91
+ delete payload.thinking
92
+ }
93
+
94
+ const rewrittenThinkingType =
95
+ disableThinkingWithTools && hasTools ? 'disabled' : undefined
96
+
97
+ const summary = summarizePayload(payload, hasTools, {
98
+ incomingThinkingType,
99
+ rewrittenThinkingType
100
+ })
101
+
102
+ const modeTag = hasTools ? '[tools]' : '[chat]'
103
+ const consoleMsg = `${modeTag} thinking.type=${String(incomingThinkingType)} -> ${
104
+ hasTools && disableThinkingWithTools ? '"disabled"' : '<deleted>'
105
+ }, model=${payload.model ?? '?'}`
106
+
107
+ return { summary, consoleMsg }
108
+ }
109
+
110
+ // ---- Create and start ----
111
+
112
+ const { start } = createProxy({
113
+ upstreamUrl,
114
+ port,
115
+ logPath,
116
+ label: 'mimo-proxy',
117
+ healthCheckExtras: { disableThinkingWithTools },
118
+ rewriteRequest: rewriteMiMo,
119
+ startupMessages: (_port, _upstreamUrl) => [
120
+ `[mimo-proxy] listening on http://127.0.0.1:${_port}/v1/chat/completions`,
121
+ `[mimo-proxy] forwarding to ${_upstreamUrl}`,
122
+ `[mimo-proxy] disable thinking with tools=${disableThinkingWithTools}`,
123
+ `[mimo-proxy] writing redacted request summaries to ${logPath}`
124
+ ]
125
+ })
126
+
127
+ start()