pi-opencode-go-provider 1.0.5 → 1.0.7

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.
Files changed (8) hide show
  1. package/AGENTS.md +1 -0
  2. package/README.md +8 -4
  3. package/deprecated-models.json +1 -0
  4. package/index.ts +34 -1
  5. package/models.json +143 -68
  6. package/package.json +1 -1
  7. package/patch.json +152 -8
  8. package/scripts/update-models.js +107 -2
package/AGENTS.md CHANGED
@@ -7,6 +7,7 @@ The following files are **idempotent** and regenerated by `scripts/update-models
7
7
  | File | Why it's auto-generated |
8
8
  |------|------------------------|
9
9
  | `models.json` | Built from the provider API. `update-models.js` fetches models, preserves curated data for known IDs, and writes this file. |
10
+ | `deprecated-models.json` | Graveyard for models the API delisted. update-models.js stamps them with deprecatedAt and pi keeps serving them for a 2-week grace period, then evicts them. Never edit by hand. |
10
11
  | `README.md` (model table) | The table under `## Available Models` is replaced in-place by `update-models.js` after merging base models → patch → custom models. |
11
12
 
12
13
  ## Correct Files to Edit
package/README.md CHANGED
@@ -69,16 +69,20 @@ pi
69
69
  | Model | API | Type | Context | Max Tokens | Input Cost | Output Cost |
70
70
  |-------|-----|------|---------|------------|------------|-------------|
71
71
  | DeepSeek V4 Flash | Completions | Text | 1.0M | 384K | $0.14 | $0.28 |
72
- | DeepSeek V4 Pro | Completions | Text | 1.0M | 384K | $1.74 | $3.48 |
72
+ | DeepSeek V4 Pro | Completions | Text | 1.0M | 384K | $0.43 | $0.87 |
73
73
  | GLM-5.1 | Completions | Text | 203K | 33K | $1.40 | $4.40 |
74
74
  | GLM-5.2 | Completions | Text | 1.0M | 131K | $1.40 | $4.40 |
75
+ | GPT-5.6 Luna (2x usage) | Completions | Text + Image | 1.1M | 128K | $0.10 | $0.60 |
76
+ | Grok 4.5 | Responses | Text + Image | 500K | 500K | $2.00 | $6.00 |
77
+ | Hy3 | Completions | Text | 256K | 64K | $0.14 | $0.58 |
75
78
  | Kimi K2.6 | Completions | Text + Image | 262K | 66K | $0.95 | $4.00 |
76
79
  | Kimi K2.7 Code | Completions | Text + Image | 262K | 262K | $0.95 | $4.00 |
80
+ | Kimi K3 (2x usage) | Completions | Text + Image | 1.0M | 131K | $3.00 | $15.00 |
77
81
  | MiMo V2.5 | Completions | Text + Image | 1.0M | 128K | $0.14 | $0.28 |
78
- | MiMo V2.5 Pro | Completions | Text | 1.0M | 128K | $1.74 | $3.48 |
79
- | MiniMax-M2.7 | Anthropic | Text | 205K | 131K | $0.30 | $1.20 |
82
+ | MiMo V2.5 Pro | Completions | Text | 1.0M | 128K | $0.43 | $0.87 |
83
+ | MiniMax-M2.7 | Completions | Text | 205K | 131K | $0.30 | $1.20 |
80
84
  | MiniMax-M3 | Anthropic | Text + Image | 1.0M | 131K | $0.30 | $1.20 |
81
- | Qwen3.6 Plus | Anthropic | Text + Image | 1.0M | 66K | $0.50 | $3.00 |
85
+ | Qwen3.6 Plus | Completions | Text + Image | 1.0M | 66K | $0.50 | $3.00 |
82
86
  | Qwen3.7 Max | Anthropic | Text | 1.0M | 66K | $2.50 | $7.50 |
83
87
  | Qwen3.7 Plus | Anthropic | Text + Image | 1.0M | 66K | $0.40 | $1.60 |
84
88
  *Costs are per million tokens. Prices subject to change - check [opencode.ai](https://opencode.ai) for current pricing.*
package/deprecated-models.json ADDED
@@ -0,0 +1 @@
1
+ {}
package/index.ts CHANGED
@@ -26,6 +26,7 @@ import { getAgentDir, type ExtensionAPI, type ModelRegistry } from "@earendil-wo
26
26
  import modelsData from "./models.json" with { type: "json" };
27
27
  import customModelsData from "./custom-models.json" with { type: "json" };
28
28
  import patchData from "./patch.json" with { type: "json" };
29
+ import deprecatedData from "./deprecated-models.json" with { type: "json" };
29
30
  import fs from "fs";
30
31
  import path from "path";
31
32
 
@@ -116,7 +117,10 @@ function applyPatch(model: JsonModel, patch: PatchEntry): JsonModel {
116
117
  function buildModels(base: JsonModel[], custom: JsonModel[], patch: PatchData): JsonModel[] {
117
118
  const modelMap = new Map<string, JsonModel>();
118
119
 
119
- for (const model of base) {
120
+ // Seed with the base list plus grace-period deprecated models so patch.json
121
+ // entries apply to deprecated models exactly as while the model was live
122
+ // (withDeprecated keeps live data on id conflicts).
123
+ for (const model of withDeprecated(base)) {
120
124
  modelMap.set(model.id, model);
121
125
  }
122
126
 
@@ -257,6 +261,35 @@ function mergeWithEmbedded(liveModels: JsonModel[], embeddedModels: JsonModel[])
257
261
  return result;
258
262
  }
259
263
 
264
+ // Grace period for delisted models. When the provider API stops listing a
265
+ // model, update-models.js moves its last-known definition into
266
+ // deprecated-models.json (stamped with deprecatedAt) instead of dropping it.
267
+ // For 14 days the model keeps working here so in-flight sessions and saved
268
+ // model settings do not break; afterwards it is evicted permanently.
269
+ const DEPRECATED_MODEL_TTL_MS = 14 * 24 * 60 * 60 * 1000;
270
+
271
+ // Grace-period deprecated models with deprecation metadata stripped.
272
+ function activeDeprecatedModels(): JsonModel[] {
273
+ const now = Date.now();
274
+ const result: JsonModel[] = [];
275
+ for (const entry of Object.values(deprecatedData as Record<string, JsonModel & { deprecatedAt?: string }>)) {
276
+ if (!entry?.id) continue;
277
+ const removedAt = Date.parse(entry.deprecatedAt ?? "");
278
+ if (Number.isNaN(removedAt) || now - removedAt > DEPRECATED_MODEL_TTL_MS) continue;
279
+ const model = { ...entry } as JsonModel & { deprecatedAt?: string };
280
+ delete model.deprecatedAt;
281
+ result.push(model);
282
+ }
283
+ return result;
284
+ }
285
+
286
+ // Append grace-period deprecated models the list does not already have (live data wins).
287
+ function withDeprecated(models: JsonModel[]): JsonModel[] {
288
+ const seen = new Set(models.map((m) => m.id));
289
+ const extras = activeDeprecatedModels().filter((m) => !seen.has(m.id));
290
+ return extras.length > 0 ? [...models, ...extras] : models;
291
+ }
292
+
260
293
  function loadStaleModels(embeddedModels: JsonModel[]): JsonModel[] {
261
294
  const cached = loadCachedModels();
262
295
  if (!cached || cached.length === 0) return embeddedModels;
package/models.json CHANGED
@@ -1,7 +1,44 @@
1
1
  [
2
+ {
3
+ "id": "qwen3.7-plus",
4
+ "name": "Qwen3.7 Plus",
5
+ "api": "anthropic-messages",
6
+ "baseUrl": "https://opencode.ai/zen/go",
7
+ "reasoning": true,
8
+ "input": [
9
+ "text",
10
+ "image"
11
+ ],
12
+ "cost": {
13
+ "input": 0.4,
14
+ "output": 1.6,
15
+ "cacheRead": 0.04,
16
+ "cacheWrite": 0.5
17
+ },
18
+ "contextWindow": 1000000,
19
+ "maxTokens": 65536
20
+ },
21
+ {
22
+ "id": "glm-5.1",
23
+ "name": "GLM-5.1",
24
+ "api": "openai-completions",
25
+ "baseUrl": "https://opencode.ai/zen/go/v1",
26
+ "reasoning": true,
27
+ "input": [
28
+ "text"
29
+ ],
30
+ "cost": {
31
+ "input": 1.4,
32
+ "output": 4.4,
33
+ "cacheRead": 0.26,
34
+ "cacheWrite": 0
35
+ },
36
+ "contextWindow": 202752,
37
+ "maxTokens": 32768
38
+ },
2
39
  {
3
40
  "id": "deepseek-v4-flash",
4
- "name": "DeepSeek V4 Flash",
41
+ "name": "DeepSeek V4 Flash (New)",
5
42
  "api": "openai-completions",
6
43
  "baseUrl": "https://opencode.ai/zen/go/v1",
7
44
  "reasoning": true,
@@ -18,23 +55,40 @@
18
55
  "maxTokens": 384000
19
56
  },
20
57
  {
21
- "id": "qwen3.7-plus",
22
- "name": "Qwen3.7 Plus",
58
+ "id": "minimax-m2.7",
59
+ "name": "MiniMax-M2.7",
23
60
  "api": "anthropic-messages",
24
61
  "baseUrl": "https://opencode.ai/zen/go",
25
62
  "reasoning": true,
26
63
  "input": [
27
- "text",
28
- "image"
64
+ "text"
29
65
  ],
30
66
  "cost": {
31
- "input": 0.4,
32
- "output": 1.6,
33
- "cacheRead": 0.04,
34
- "cacheWrite": 0.5
67
+ "input": 0.3,
68
+ "output": 1.2,
69
+ "cacheRead": 0.06,
70
+ "cacheWrite": 0
71
+ },
72
+ "contextWindow": 204800,
73
+ "maxTokens": 131072
74
+ },
75
+ {
76
+ "id": "glm-5.2",
77
+ "name": "GLM-5.2",
78
+ "api": "openai-completions",
79
+ "baseUrl": "https://opencode.ai/zen/go/v1",
80
+ "reasoning": true,
81
+ "input": [
82
+ "text"
83
+ ],
84
+ "cost": {
85
+ "input": 1.4,
86
+ "output": 4.4,
87
+ "cacheRead": 0.26,
88
+ "cacheWrite": 0
35
89
  },
36
90
  "contextWindow": 1000000,
37
- "maxTokens": 65536
91
+ "maxTokens": 131072
38
92
  },
39
93
  {
40
94
  "id": "qwen3.7-max",
@@ -55,8 +109,8 @@
55
109
  "maxTokens": 65536
56
110
  },
57
111
  {
58
- "id": "kimi-k2.7-code",
59
- "name": "Kimi K2.7 Code",
112
+ "id": "kimi-k2.6",
113
+ "name": "Kimi K2.6",
60
114
  "api": "openai-completions",
61
115
  "baseUrl": "https://opencode.ai/zen/go/v1",
62
116
  "reasoning": true,
@@ -67,15 +121,34 @@
67
121
  "cost": {
68
122
  "input": 0.95,
69
123
  "output": 4,
70
- "cacheRead": 0.19,
124
+ "cacheRead": 0.16,
71
125
  "cacheWrite": 0
72
126
  },
73
127
  "contextWindow": 262144,
74
- "maxTokens": 262144
128
+ "maxTokens": 65536
75
129
  },
76
130
  {
77
- "id": "glm-5.1",
78
- "name": "GLM-5.1",
131
+ "id": "minimax-m3",
132
+ "name": "MiniMax-M3",
133
+ "api": "anthropic-messages",
134
+ "baseUrl": "https://opencode.ai/zen/go",
135
+ "reasoning": true,
136
+ "input": [
137
+ "text",
138
+ "image"
139
+ ],
140
+ "cost": {
141
+ "input": 0.3,
142
+ "output": 1.2,
143
+ "cacheRead": 0.06,
144
+ "cacheWrite": 0
145
+ },
146
+ "contextWindow": 1000000,
147
+ "maxTokens": 131072
148
+ },
149
+ {
150
+ "id": "hy3",
151
+ "name": "Hy3",
79
152
  "api": "openai-completions",
80
153
  "baseUrl": "https://opencode.ai/zen/go/v1",
81
154
  "reasoning": true,
@@ -83,13 +156,13 @@
83
156
  "text"
84
157
  ],
85
158
  "cost": {
86
- "input": 1.4,
87
- "output": 4.4,
88
- "cacheRead": 0.26,
159
+ "input": 0.14,
160
+ "output": 0.58,
161
+ "cacheRead": 0.035,
89
162
  "cacheWrite": 0
90
163
  },
91
- "contextWindow": 202752,
92
- "maxTokens": 32768
164
+ "contextWindow": 256000,
165
+ "maxTokens": 64000
93
166
  },
94
167
  {
95
168
  "id": "deepseek-v4-pro",
@@ -101,72 +174,74 @@
101
174
  "text"
102
175
  ],
103
176
  "cost": {
104
- "input": 1.74,
105
- "output": 3.48,
106
- "cacheRead": 0.0145,
177
+ "input": 0.435,
178
+ "output": 0.87,
179
+ "cacheRead": 0.003625,
107
180
  "cacheWrite": 0
108
181
  },
109
182
  "contextWindow": 1000000,
110
183
  "maxTokens": 384000
111
184
  },
112
185
  {
113
- "id": "glm-5.2",
114
- "name": "GLM-5.2",
186
+ "id": "mimo-v2.5",
187
+ "name": "MiMo V2.5",
115
188
  "api": "openai-completions",
116
189
  "baseUrl": "https://opencode.ai/zen/go/v1",
117
190
  "reasoning": true,
118
191
  "input": [
119
- "text"
192
+ "text",
193
+ "image"
120
194
  ],
121
195
  "cost": {
122
- "input": 1.4,
123
- "output": 4.4,
124
- "cacheRead": 0.26,
196
+ "input": 0.14,
197
+ "output": 0.28,
198
+ "cacheRead": 0.0028,
125
199
  "cacheWrite": 0
126
200
  },
127
201
  "contextWindow": 1000000,
128
- "maxTokens": 131072
202
+ "maxTokens": 128000
129
203
  },
130
204
  {
131
- "id": "minimax-m3",
132
- "name": "MiniMax-M3",
133
- "api": "anthropic-messages",
134
- "baseUrl": "https://opencode.ai/zen/go",
205
+ "id": "gpt-5.6-luna",
206
+ "name": "GPT-5.6 Luna (2x usage)",
207
+ "api": "openai-completions",
208
+ "baseUrl": "https://opencode.ai/zen/go/v1",
135
209
  "reasoning": true,
136
210
  "input": [
137
211
  "text",
138
212
  "image"
139
213
  ],
140
214
  "cost": {
141
- "input": 0.3,
142
- "output": 1.2,
143
- "cacheRead": 0.06,
144
- "cacheWrite": 0
215
+ "input": 0.1,
216
+ "output": 0.6,
217
+ "cacheRead": 0.01,
218
+ "cacheWrite": 0.125
145
219
  },
146
- "contextWindow": 1000000,
147
- "maxTokens": 131072
220
+ "contextWindow": 1050000,
221
+ "maxTokens": 128000
148
222
  },
149
223
  {
150
- "id": "minimax-m2.7",
151
- "name": "MiniMax-M2.7",
152
- "api": "anthropic-messages",
153
- "baseUrl": "https://opencode.ai/zen/go",
224
+ "id": "grok-4.5",
225
+ "name": "Grok 4.5",
226
+ "api": "openai-completions",
227
+ "baseUrl": "https://opencode.ai/zen/go/v1",
154
228
  "reasoning": true,
155
229
  "input": [
156
- "text"
230
+ "text",
231
+ "image"
157
232
  ],
158
233
  "cost": {
159
- "input": 0.3,
160
- "output": 1.2,
161
- "cacheRead": 0.06,
234
+ "input": 2,
235
+ "output": 6,
236
+ "cacheRead": 0.5,
162
237
  "cacheWrite": 0
163
238
  },
164
- "contextWindow": 204800,
165
- "maxTokens": 131072
239
+ "contextWindow": 500000,
240
+ "maxTokens": 500000
166
241
  },
167
242
  {
168
- "id": "mimo-v2.5",
169
- "name": "MiMo V2.5",
243
+ "id": "kimi-k2.7-code",
244
+ "name": "Kimi K2.7 Code",
170
245
  "api": "openai-completions",
171
246
  "baseUrl": "https://opencode.ai/zen/go/v1",
172
247
  "reasoning": true,
@@ -175,17 +250,17 @@
175
250
  "image"
176
251
  ],
177
252
  "cost": {
178
- "input": 0.14,
179
- "output": 0.28,
180
- "cacheRead": 0.0028,
253
+ "input": 0.95,
254
+ "output": 4,
255
+ "cacheRead": 0.19,
181
256
  "cacheWrite": 0
182
257
  },
183
- "contextWindow": 1000000,
184
- "maxTokens": 128000
258
+ "contextWindow": 262144,
259
+ "maxTokens": 262144
185
260
  },
186
261
  {
187
- "id": "kimi-k2.6",
188
- "name": "Kimi K2.6",
262
+ "id": "kimi-k3",
263
+ "name": "Kimi K3",
189
264
  "api": "openai-completions",
190
265
  "baseUrl": "https://opencode.ai/zen/go/v1",
191
266
  "reasoning": true,
@@ -194,13 +269,13 @@
194
269
  "image"
195
270
  ],
196
271
  "cost": {
197
- "input": 0.95,
198
- "output": 4,
199
- "cacheRead": 0.16,
272
+ "input": 3,
273
+ "output": 15,
274
+ "cacheRead": 0.3,
200
275
  "cacheWrite": 0
201
276
  },
202
- "contextWindow": 262144,
203
- "maxTokens": 65536
277
+ "contextWindow": 1048576,
278
+ "maxTokens": 131072
204
279
  },
205
280
  {
206
281
  "id": "mimo-v2.5-pro",
@@ -212,9 +287,9 @@
212
287
  "text"
213
288
  ],
214
289
  "cost": {
215
- "input": 1.74,
216
- "output": 3.48,
217
- "cacheRead": 0.0145,
290
+ "input": 0.435,
291
+ "output": 0.87,
292
+ "cacheRead": 0.003625,
218
293
  "cacheWrite": 0
219
294
  },
220
295
  "contextWindow": 1048576,
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "pi-opencode-go-provider",
3
- "version": "1.0.5",
3
+ "version": "1.0.7",
4
4
  "description": "Opencode Go provider extension for pi - Fast, efficient GLM, Kimi, and MiniMax models through the opencode.ai API",
5
5
  "type": "module",
6
6
  "main": "index.ts",
package/patch.json CHANGED
@@ -1,17 +1,161 @@
1
1
  {
2
+ "glm-5.1": {
3
+ "compat": {
4
+ "supportsStore": false,
5
+ "supportsDeveloperRole": false,
6
+ "maxTokensField": "max_tokens"
7
+ }
8
+ },
2
9
  "deepseek-v4-flash": {
3
- "compat": { "requiresReasoningContentOnAssistantMessages": true, "thinkingFormat": "deepseek" },
4
- "thinkingLevelMap": { "minimal": null, "low": null, "medium": null, "high": "high", "max": "max" }
10
+ "name": "DeepSeek V4 Flash",
11
+ "thinkingLevelMap": {
12
+ "minimal": null,
13
+ "low": null,
14
+ "medium": null,
15
+ "high": "high",
16
+ "max": "max"
17
+ },
18
+ "compat": {
19
+ "supportsStore": false,
20
+ "supportsDeveloperRole": false,
21
+ "maxTokensField": "max_tokens",
22
+ "requiresReasoningContentOnAssistantMessages": true,
23
+ "thinkingFormat": "deepseek"
24
+ }
5
25
  },
6
- "deepseek-v4-pro": {
7
- "compat": { "requiresReasoningContentOnAssistantMessages": true, "thinkingFormat": "deepseek" },
8
- "thinkingLevelMap": { "minimal": null, "low": null, "medium": null, "high": "high", "max": "max" }
26
+ "minimax-m2.7": {
27
+ "api": "openai-completions",
28
+ "baseUrl": "https://opencode.ai/zen/go/v1",
29
+ "compat": {
30
+ "supportsStore": false,
31
+ "supportsDeveloperRole": false,
32
+ "maxTokensField": "max_tokens"
33
+ }
34
+ },
35
+ "glm-5.2": {
36
+ "thinkingLevelMap": {
37
+ "off": null,
38
+ "minimal": null,
39
+ "low": null,
40
+ "medium": null,
41
+ "high": "high",
42
+ "xhigh": null,
43
+ "max": "max"
44
+ },
45
+ "compat": {
46
+ "supportsStore": false,
47
+ "supportsDeveloperRole": false,
48
+ "maxTokensField": "max_tokens"
49
+ }
9
50
  },
10
51
  "kimi-k2.6": {
11
- "compat": { "thinkingFormat": "deepseek", "supportsReasoningEffort": false },
12
- "thinkingLevelMap": { "minimal": null, "low": null, "medium": null }
52
+ "thinkingLevelMap": {
53
+ "minimal": null,
54
+ "low": null,
55
+ "medium": null
56
+ },
57
+ "compat": {
58
+ "supportsStore": false,
59
+ "supportsDeveloperRole": false,
60
+ "thinkingFormat": "deepseek",
61
+ "supportsReasoningEffort": false,
62
+ "maxTokensField": "max_tokens",
63
+ "supportsLongCacheRetention": false
64
+ }
65
+ },
66
+ "hy3": {
67
+ "thinkingLevelMap": {
68
+ "off": "none",
69
+ "minimal": null,
70
+ "low": "low",
71
+ "medium": null,
72
+ "high": "high",
73
+ "xhigh": null,
74
+ "max": null
75
+ },
76
+ "compat": {
77
+ "supportsStore": false,
78
+ "supportsDeveloperRole": false,
79
+ "maxTokensField": "max_tokens"
80
+ }
81
+ },
82
+ "deepseek-v4-pro": {
83
+ "thinkingLevelMap": {
84
+ "minimal": null,
85
+ "low": null,
86
+ "medium": null,
87
+ "high": "high",
88
+ "max": "max"
89
+ },
90
+ "compat": {
91
+ "supportsStore": false,
92
+ "supportsDeveloperRole": false,
93
+ "maxTokensField": "max_tokens",
94
+ "requiresReasoningContentOnAssistantMessages": true,
95
+ "thinkingFormat": "deepseek"
96
+ }
97
+ },
98
+ "mimo-v2.5": {
99
+ "compat": {
100
+ "supportsStore": false,
101
+ "supportsDeveloperRole": false,
102
+ "maxTokensField": "max_tokens"
103
+ }
104
+ },
105
+ "grok-4.5": {
106
+ "api": "openai-responses",
107
+ "thinkingLevelMap": {
108
+ "off": null,
109
+ "minimal": null,
110
+ "low": "low",
111
+ "medium": "medium",
112
+ "high": "high",
113
+ "xhigh": null,
114
+ "max": null
115
+ },
116
+ "compat": {
117
+ "sessionAffinityFormat": "openai-nosession"
118
+ }
119
+ },
120
+ "kimi-k2.7-code": {
121
+ "compat": {
122
+ "supportsStore": false,
123
+ "supportsDeveloperRole": false,
124
+ "maxTokensField": "max_tokens"
125
+ }
126
+ },
127
+ "kimi-k3": {
128
+ "name": "Kimi K3 (2x usage)",
129
+ "thinkingLevelMap": {
130
+ "off": null,
131
+ "minimal": null,
132
+ "low": null,
133
+ "medium": null,
134
+ "high": null,
135
+ "xhigh": null,
136
+ "max": "max"
137
+ },
138
+ "compat": {
139
+ "supportsStore": false,
140
+ "supportsDeveloperRole": false,
141
+ "maxTokensField": "max_tokens"
142
+ }
143
+ },
144
+ "mimo-v2.5-pro": {
145
+ "compat": {
146
+ "supportsStore": false,
147
+ "supportsDeveloperRole": false,
148
+ "maxTokensField": "max_tokens"
149
+ }
13
150
  },
14
151
  "qwen3.6-plus": {
15
- "compat": { "thinkingFormat": "qwen" }
152
+ "api": "openai-completions",
153
+ "baseUrl": "https://opencode.ai/zen/go/v1",
154
+ "compat": {
155
+ "supportsStore": false,
156
+ "supportsDeveloperRole": false,
157
+ "thinkingFormat": "qwen",
158
+ "maxTokensField": "max_tokens"
159
+ }
16
160
  }
17
161
  }
package/scripts/update-models.js CHANGED
@@ -101,6 +101,46 @@ function convertModel(model) {
101
101
  };
102
102
  }
103
103
 
104
+ // Load a layered JSON model file.
105
+ function loadJson(fileName) {
106
+ try {
107
+ return JSON.parse(fs.readFileSync(path.join(process.cwd(), fileName), 'utf8'));
108
+ } catch {
109
+ return fileName === 'patch.json' ? {} : [];
110
+ }
111
+ }
112
+
113
+ // Deep-merge patch overrides into a model for README documentation. The
114
+ // generated models.json remains API-derived; only the displayed model map uses patches.
115
+ function applyPatch(model, patch) {
116
+ if (!patch) return model;
117
+ const result = { ...model };
118
+ for (const [key, value] of Object.entries(patch)) {
119
+ if (key === 'cost' || key === 'compat' || key === 'thinkingLevelMap') {
120
+ result[key] = { ...(result[key] || {}), ...value };
121
+ } else {
122
+ result[key] = value;
123
+ }
124
+ }
125
+ if (!result.reasoning) {
126
+ delete result.thinkingLevelMap;
127
+ if (result.compat?.thinkingFormat) delete result.compat.thinkingFormat;
128
+ }
129
+ return result;
130
+ }
131
+
132
+ // Merge API models, patches, and custom models in the same order as index.ts.
133
+ function buildModels(baseModels, customModels, patch) {
134
+ const byId = new Map(baseModels.map(model => [model.id, model]));
135
+ for (const [id, entry] of Object.entries(patch)) {
136
+ if (byId.has(id)) byId.set(id, applyPatch(byId.get(id), entry));
137
+ }
138
+ for (const model of customModels) {
139
+ byId.set(model.id, applyPatch(model, patch[model.id]));
140
+ }
141
+ return Array.from(byId.values());
142
+ }
143
+
104
144
  // Generate README model table row
105
145
  function generateReadmeRow(model) {
106
146
  const cost = model.cost || {};
@@ -149,6 +189,67 @@ ${tableRows}`;
149
189
  console.log(` Updated README.md with ${models.length} models`);
150
190
  }
151
191
 
192
+ // Grace period for delisted models: update-models.js moves models the API no
193
+ // longer lists into deprecated-models.json (stamped with deprecatedAt) instead
194
+ // of dropping them; the runtime appends them back so sessions and saved model
195
+ // settings keep working, and after 14 days they are evicted permanently.
196
+ const DEPRECATED_MODEL_TTL_MS = 14 * 24 * 60 * 60 * 1000;
197
+
198
+ /**
199
+ * Reconcile deprecated-models.json against the freshly fetched model list.
200
+ * - in old models.json but not the API: moved into the deprecated file
201
+ * (deprecatedAt = now; preserved on repeat runs so the grace clock is not reset)
202
+ * - back in the API: resurrected (dropped from the deprecated file)
203
+ * - deprecatedAt older than 14 days: evicted permanently
204
+ * Must run BEFORE the new models.json is written; it reads the old file itself.
205
+ */
206
+ function updateDeprecatedModels(modelsJsonPath, newModels) {
207
+ const deprecatedPath = path.join(path.dirname(modelsJsonPath), 'deprecated-models.json');
208
+
209
+ let oldModels = [];
210
+ try {
211
+ const parsed = JSON.parse(fs.readFileSync(modelsJsonPath, 'utf8'));
212
+ if (Array.isArray(parsed)) oldModels = parsed;
213
+ } catch { /* first run: no previous models.json */ }
214
+
215
+ let deprecated = {};
216
+ try {
217
+ const parsed = JSON.parse(fs.readFileSync(deprecatedPath, 'utf8'));
218
+ if (parsed && typeof parsed === 'object' && !Array.isArray(parsed)) deprecated = parsed;
219
+ } catch { /* no graveyard yet */ }
220
+
221
+ const currentIds = new Set(newModels.map(m => m.id));
222
+ const now = new Date().toISOString();
223
+ const added = [];
224
+ const resurrected = [];
225
+ const evicted = [];
226
+
227
+ for (const old of oldModels) {
228
+ if (old && old.id && !currentIds.has(old.id) && !deprecated[old.id]) {
229
+ deprecated[old.id] = { ...old, deprecatedAt: now };
230
+ added.push(old.id);
231
+ }
232
+ }
233
+
234
+ for (const [id, entry] of Object.entries(deprecated)) {
235
+ if (currentIds.has(id)) {
236
+ delete deprecated[id];
237
+ resurrected.push(id);
238
+ continue;
239
+ }
240
+ const removedAt = Date.parse(entry && entry.deprecatedAt ? entry.deprecatedAt : '');
241
+ if (Number.isNaN(removedAt) || Date.now() - removedAt > DEPRECATED_MODEL_TTL_MS) {
242
+ delete deprecated[id];
243
+ evicted.push(id);
244
+ }
245
+ }
246
+
247
+ if (added.length > 0 || resurrected.length > 0 || evicted.length > 0) {
248
+ fs.writeFileSync(deprecatedPath, JSON.stringify(deprecated, null, 2) + '\n');
249
+ console.log('Updated deprecated-models.json ' + JSON.stringify({ added, resurrected, evicted }));
250
+ }
251
+ }
252
+
152
253
  async function main() {
153
254
  console.log('Fetching models from API...');
154
255
 
@@ -172,11 +273,15 @@ async function main() {
172
273
  // Convert to Pi-native format and save to models.json
173
274
  const models = apiModels.map(convertModel);
174
275
  const modelsPath = path.join(process.cwd(), 'models.json');
276
+ // Move delisted models to deprecated-models.json BEFORE models.json is overwritten
277
+ updateDeprecatedModels(modelsPath, models);
175
278
  fs.writeFileSync(modelsPath, JSON.stringify(models, null, 2) + '\n');
176
279
  console.log(` Saved ${models.length} models to models.json`);
177
280
 
178
- // Update README
179
- updateReadme(models);
281
+ // Apply layered overrides for documentation: models.json → patch.json → custom-models.json.
282
+ const patch = loadJson('patch.json');
283
+ const customModels = Array.isArray(loadJson('custom-models.json')) ? loadJson('custom-models.json') : [];
284
+ updateReadme(buildModels(models, customModels, patch));
180
285
 
181
286
  console.log('\nDone!');
182
287
  } catch (error) {