@genesislcap/foundation-ai 15.90.4 → 15.92.0
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 +26 -103
- package/dist/dts/ai-provider.d.ts +5 -3
- package/dist/dts/ai-provider.d.ts.map +1 -1
- package/dist/dts/index.d.ts +7 -4
- package/dist/dts/index.d.ts.map +1 -1
- package/dist/dts/providers/openai-provider.d.ts +8 -7
- package/dist/dts/providers/openai-provider.d.ts.map +1 -1
- package/dist/dts/transports/budget-exhausted-error.d.ts +5 -7
- package/dist/dts/transports/budget-exhausted-error.d.ts.map +1 -1
- package/dist/dts/transports/context-overflow-error.d.ts.map +1 -1
- package/dist/dts/transports/openai-transport.d.ts +119 -0
- package/dist/dts/transports/openai-transport.d.ts.map +1 -0
- package/dist/dts/types/chat.types.d.ts +12 -0
- package/dist/dts/types/chat.types.d.ts.map +1 -1
- package/dist/dts/types/config.types.d.ts +44 -16
- package/dist/dts/types/config.types.d.ts.map +1 -1
- package/dist/dts/utils/token-cost.d.ts +65 -1
- package/dist/dts/utils/token-cost.d.ts.map +1 -1
- package/dist/dts/utils/tool-schema.d.ts +38 -0
- package/dist/dts/utils/tool-schema.d.ts.map +1 -1
- package/dist/esm/ai-provider.js +21 -14
- package/dist/esm/index.js +4 -2
- package/dist/esm/providers/openai-provider.js +15 -9
- package/dist/esm/transports/budget-exhausted-error.js +6 -7
- package/dist/esm/transports/context-overflow-error.js +2 -0
- package/dist/esm/transports/openai-transport.js +752 -0
- package/dist/esm/types/config.types.js +12 -1
- package/dist/esm/utils/token-cost.js +89 -0
- package/dist/esm/utils/tool-schema.js +108 -0
- package/dist/foundation-ai.api.json +1745 -137
- package/dist/foundation-ai.d.ts +270 -28
- package/package.json +9 -9
- package/dist/dts/transports/server-openai-transport.d.ts +0 -23
- package/dist/dts/transports/server-openai-transport.d.ts.map +0 -1
- package/dist/esm/transports/server-openai-transport.js +0 -94
|
@@ -20,11 +20,26 @@ export declare const SUPPORTED_GEMINI_MODEL_IDS: readonly GeminiModelId[];
|
|
|
20
20
|
export type AnthropicModelId = 'claude-fable-5' | 'claude-opus-4-8' | 'claude-opus-4-7' | 'claude-sonnet-5' | 'claude-sonnet-4-6' | 'claude-haiku-4-5-20251001';
|
|
21
21
|
/** @beta */
|
|
22
22
|
export declare const SUPPORTED_ANTHROPIC_MODEL_IDS: readonly AnthropicModelId[];
|
|
23
|
+
/**
|
|
24
|
+
* OpenAI Responses API model ids supported by {@link OpenAITransport}.
|
|
25
|
+
*
|
|
26
|
+
* @remarks
|
|
27
|
+
* The GPT-6 family only: on gpt-6-astra, gpt-6.1-sol and gpt-6.1-terra, tool calling is available
|
|
28
|
+
* on the Responses API alone, and the transport is built on it. Older ids are not listed because
|
|
29
|
+
* nothing here prices or tests them — see {@link SUPPORTED_ANTHROPIC_MODEL_IDS} for why an unlisted
|
|
30
|
+
* model must be a decision rather than a fall-through.
|
|
31
|
+
*
|
|
32
|
+
* @beta
|
|
33
|
+
*/
|
|
34
|
+
export type OpenAIModelId = 'gpt-6-astra' | 'gpt-6.1-sol' | 'gpt-6.1-terra' | 'gpt-6-sol' | 'gpt-6-luna';
|
|
35
|
+
/** @beta */
|
|
36
|
+
export declare const SUPPORTED_OPENAI_MODEL_IDS: readonly OpenAIModelId[];
|
|
23
37
|
/**
|
|
24
38
|
* Which vendor a model id belongs to, or `undefined` when no supported provider claims it.
|
|
25
39
|
*
|
|
26
40
|
* @remarks
|
|
27
|
-
* Derived from {@link SUPPORTED_ANTHROPIC_MODEL_IDS}
|
|
41
|
+
* Derived from {@link SUPPORTED_ANTHROPIC_MODEL_IDS}, {@link SUPPORTED_GEMINI_MODEL_IDS} and
|
|
42
|
+
* {@link SUPPORTED_OPENAI_MODEL_IDS},
|
|
28
43
|
* so it cannot drift from them the way a hand-maintained map does — adding a model to an
|
|
29
44
|
* allowlist is enough to teach this too.
|
|
30
45
|
*
|
|
@@ -68,20 +83,6 @@ export interface AIProviderConfig {
|
|
|
68
83
|
/** Optional override for criteria interpretation capability. */
|
|
69
84
|
criteriaInstructions?: string;
|
|
70
85
|
}
|
|
71
|
-
/** Server-proxy providers only (API keys stay on server). */
|
|
72
|
-
export type ServerAIProviderType = Extract<AIProviderType, 'openai'>;
|
|
73
|
-
/**
|
|
74
|
-
* Server-proxy AI configuration (client calls your server; server calls the provider).
|
|
75
|
-
*
|
|
76
|
-
* @remarks
|
|
77
|
-
* API key stays on the server. URL is derived from API_HOST at /ai-service/chat-completions.
|
|
78
|
-
* Use provider to select backend (openai, anthropic, etc.).
|
|
79
|
-
*
|
|
80
|
-
* @beta
|
|
81
|
-
*/
|
|
82
|
-
export interface ServerAIConfig extends AIProviderConfig {
|
|
83
|
-
provider: ServerAIProviderType;
|
|
84
|
-
}
|
|
85
86
|
/**
|
|
86
87
|
* Chrome built-in AI (Prompt API / Gemini Nano) configuration.
|
|
87
88
|
*
|
|
@@ -133,11 +134,38 @@ export interface AnthropicAIConfig extends AIProviderConfig {
|
|
|
133
134
|
*/
|
|
134
135
|
maxTokens?: number;
|
|
135
136
|
}
|
|
137
|
+
/**
|
|
138
|
+
* OpenAI server-proxy AI configuration (client calls your server; server calls OpenAI's Responses
|
|
139
|
+
* API).
|
|
140
|
+
*
|
|
141
|
+
* @remarks
|
|
142
|
+
* API key stays on the server: set `serverEndpoint` to the proxy route (Create's ai-service mounts
|
|
143
|
+
* it at `/api/openai/chat`, e.g. `/gwf/ai-service/api/openai/chat`). There is no default, and with
|
|
144
|
+
* neither `serverEndpoint` nor `apiKey` the transport throws. When `apiKey` is provided the
|
|
145
|
+
* transport calls `api.openai.com/v1/responses` directly — only use this in trusted contexts.
|
|
146
|
+
*
|
|
147
|
+
* @beta
|
|
148
|
+
*/
|
|
149
|
+
export interface OpenAIAIConfig extends AIProviderConfig {
|
|
150
|
+
providerType: 'openai';
|
|
151
|
+
/** Defaults to `gpt-6-luna`. Only {@link SUPPORTED_OPENAI_MODEL_IDS} are accepted at runtime. */
|
|
152
|
+
model?: OpenAIModelId;
|
|
153
|
+
serverEndpoint?: string;
|
|
154
|
+
apiKey?: string;
|
|
155
|
+
/** Per-response output ceiling (`max_output_tokens`). Omitted when unset. */
|
|
156
|
+
maxTokens?: number;
|
|
157
|
+
/**
|
|
158
|
+
* Ask for reasoning summaries (`reasoning.summary`) so a thinking turn has something to show.
|
|
159
|
+
* Off by default: OpenAI may require organisation verification before it returns them, and an
|
|
160
|
+
* unverified organisation would otherwise see every reasoning turn rejected.
|
|
161
|
+
*/
|
|
162
|
+
reasoningSummary?: 'auto' | 'concise' | 'detailed';
|
|
163
|
+
}
|
|
136
164
|
/**
|
|
137
165
|
* Union type for provider-specific configs.
|
|
138
166
|
* Extend when adding new providers.
|
|
139
167
|
*
|
|
140
168
|
* @beta
|
|
141
169
|
*/
|
|
142
|
-
export type AIConfig =
|
|
170
|
+
export type AIConfig = ChromeAIConfig | GeminiAIConfig | AnthropicAIConfig | OpenAIAIConfig;
|
|
143
171
|
//# sourceMappingURL=config.types.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"config.types.d.ts","sourceRoot":"","sources":["../../../src/types/config.types.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,MAAM,MAAM,cAAc,GAAG,QAAQ,GAAG,QAAQ,GAAG,QAAQ,GAAG,WAAW,GAAG,MAAM,CAAC;AAEnF;;;;GAIG;AACH,MAAM,MAAM,aAAa,GACrB,gBAAgB,GAChB,kBAAkB,GAClB,uBAAuB,GACvB,gBAAgB,GAChB,kBAAkB,GAClB,kBAAkB,GAClB,uBAAuB,GAEvB,wBAAwB,CAAC;AAE7B,YAAY;AACZ,eAAO,MAAM,0BAA0B,EAAE,SAAS,aAAa,EASrD,CAAC;AAEX;;;;GAIG;AACH,MAAM,MAAM,gBAAgB,GACxB,gBAAgB,GAChB,iBAAiB,GACjB,iBAAiB,GACjB,iBAAiB,GACjB,mBAAmB,GACnB,2BAA2B,CAAC;AAEhC,YAAY;AACZ,eAAO,MAAM,6BAA6B,EAAE,SAAS,gBAAgB,EAO3D,CAAC;AAEX
|
|
1
|
+
{"version":3,"file":"config.types.d.ts","sourceRoot":"","sources":["../../../src/types/config.types.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,MAAM,MAAM,cAAc,GAAG,QAAQ,GAAG,QAAQ,GAAG,QAAQ,GAAG,WAAW,GAAG,MAAM,CAAC;AAEnF;;;;GAIG;AACH,MAAM,MAAM,aAAa,GACrB,gBAAgB,GAChB,kBAAkB,GAClB,uBAAuB,GACvB,gBAAgB,GAChB,kBAAkB,GAClB,kBAAkB,GAClB,uBAAuB,GAEvB,wBAAwB,CAAC;AAE7B,YAAY;AACZ,eAAO,MAAM,0BAA0B,EAAE,SAAS,aAAa,EASrD,CAAC;AAEX;;;;GAIG;AACH,MAAM,MAAM,gBAAgB,GACxB,gBAAgB,GAChB,iBAAiB,GACjB,iBAAiB,GACjB,iBAAiB,GACjB,mBAAmB,GACnB,2BAA2B,CAAC;AAEhC,YAAY;AACZ,eAAO,MAAM,6BAA6B,EAAE,SAAS,gBAAgB,EAO3D,CAAC;AAEX;;;;;;;;;;GAUG;AACH,MAAM,MAAM,aAAa,GACrB,aAAa,GACb,aAAa,GACb,eAAe,GACf,WAAW,GACX,YAAY,CAAC;AAEjB,YAAY;AACZ,eAAO,MAAM,0BAA0B,EAAE,SAAS,aAAa,EAMrD,CAAC;AAEX;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,wBAAgB,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS,CAKzE;AAED;;;;;;;;GAQG;AACH,MAAM,WAAW,gBAAgB;IAC/B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;;;;OAKG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,gEAAgE;IAChE,oBAAoB,CAAC,EAAE,MAAM,CAAC;CAC/B;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,cAAe,SAAQ,gBAAgB;IACtD,YAAY,CAAC,EAAE,QAAQ,CAAC;IACxB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,cAAe,SAAQ,gBAAgB;IACtD,YAAY,EAAE,QAAQ,CAAC;IACvB,4GAA4G;IAC5G,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED;;;;;;;;;;GAUG;AACH,MAAM,WAAW,iBAAkB,SAAQ,gBAAgB;IACzD,YAAY,EAAE,WAAW,CAAC;IAC1B,mHAAmH;IACnH,KAAK,CAAC,EAAE,gBAAgB,CAAC;IACzB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;;;OAGG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;;;;;;;;GAWG;AACH,MAAM,WAAW,cAAe,SAAQ,gBAAgB;IACtD,YAAY,EAAE,QAAQ,CAAC;IACvB,iGAAiG;IACjG,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6EAA6E;IAC7E,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,GAAG,UAAU,CAAC;CACpD;AAED;;;;;GAKG;AACH,MAAM,MAAM,QAAQ,GAAG,cAAc,GAAG,cAAc,GAAG,iBAAiB,GAAG,cAAc,CAAC"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import type { AnthropicModelId, GeminiModelId } from '../types';
|
|
1
|
+
import type { AnthropicModelId, GeminiModelId, OpenAIModelId } from '../types';
|
|
2
2
|
/**
|
|
3
3
|
* USD per million tokens, split by direction. `prompt` covers input/prompt tokens,
|
|
4
4
|
* `candidate` covers generated output (including any reasoning tokens the provider
|
|
@@ -223,4 +223,68 @@ export interface GeminiUsageRecord {
|
|
|
223
223
|
* @beta
|
|
224
224
|
*/
|
|
225
225
|
export declare function geminiTokenCost(model: GeminiModelId, usage: GeminiUsageRecord, at?: Date): TokenCost;
|
|
226
|
+
/**
|
|
227
|
+
* Above this many input tokens the WHOLE request bills at the long-context tier — input and cached
|
|
228
|
+
* input at {@link OPENAI_LONG_CONTEXT_INPUT_MULTIPLIER}, output at
|
|
229
|
+
* {@link OPENAI_LONG_CONTEXT_OUTPUT_MULTIPLIER}. Like Gemini's tier it is a property of the request,
|
|
230
|
+
* not a marginal rate, and it is chosen on the whole prompt: uncached, cached and written tokens.
|
|
231
|
+
*
|
|
232
|
+
* @beta
|
|
233
|
+
*/
|
|
234
|
+
export declare const OPENAI_LONG_CONTEXT_THRESHOLD = 272000;
|
|
235
|
+
/** @beta */
|
|
236
|
+
export declare const OPENAI_LONG_CONTEXT_INPUT_MULTIPLIER = 2;
|
|
237
|
+
/** @beta */
|
|
238
|
+
export declare const OPENAI_LONG_CONTEXT_OUTPUT_MULTIPLIER = 1.5;
|
|
239
|
+
/**
|
|
240
|
+
* Cache-write multipliers on the base input rate, by TTL —
|
|
241
|
+
* https://developers.openai.com/api/docs/guides/prompt-caching. These are the WHOLE rate for a
|
|
242
|
+
* written token, not a premium added to it, which is the same shape as Anthropic's.
|
|
243
|
+
*
|
|
244
|
+
* @beta
|
|
245
|
+
*/
|
|
246
|
+
export declare const OPENAI_CACHE_WRITE_30M_MULTIPLIER = 1.25;
|
|
247
|
+
/** @beta */
|
|
248
|
+
export declare const OPENAI_CACHE_WRITE_12H_MULTIPLIER = 2.5;
|
|
249
|
+
/**
|
|
250
|
+
* The rates a request on `model` with `promptTokens` of input is billed at. `promptTokens` is the
|
|
251
|
+
* whole prompt; pass 0 when only the standard tier is wanted.
|
|
252
|
+
*
|
|
253
|
+
* @beta
|
|
254
|
+
*/
|
|
255
|
+
export declare function openaiRatesFor(model: OpenAIModelId, promptTokens: number): TokenRates & {
|
|
256
|
+
cachedPerMillion: number;
|
|
257
|
+
};
|
|
258
|
+
/**
|
|
259
|
+
* Token counts from an OpenAI Responses `usage` block.
|
|
260
|
+
*
|
|
261
|
+
* @remarks
|
|
262
|
+
* Gemini's input semantics, not Anthropic's: `input_tokens` is the **whole** prompt and both cache
|
|
263
|
+
* buckets sit inside it, so the uncached remainder is a subtraction. Reasoning tokens need no field
|
|
264
|
+
* of their own — `output_tokens` already includes `output_tokens_details.reasoning_tokens`.
|
|
265
|
+
*
|
|
266
|
+
* @beta
|
|
267
|
+
*/
|
|
268
|
+
export interface OpenAIUsageRecord {
|
|
269
|
+
/** `usage.input_tokens` — the whole prompt, cached reads AND writes included. */
|
|
270
|
+
inputTokens: number;
|
|
271
|
+
/** `usage.output_tokens`. Includes reasoning tokens, which bill as output. */
|
|
272
|
+
outputTokens: number;
|
|
273
|
+
/** `usage.input_tokens_details.cached_tokens` — cache reads. */
|
|
274
|
+
cacheReadTokens: number;
|
|
275
|
+
/** `usage.input_tokens_details.cache_write_tokens` — the total across both TTLs. */
|
|
276
|
+
cacheWriteTokens: number;
|
|
277
|
+
/**
|
|
278
|
+
* `usage.input_tokens_details.cache_write_tokens_details.ttl_12h_tokens` — the 12-hour
|
|
279
|
+
* portion of {@link OpenAIUsageRecord.cacheWriteTokens}; the remainder is costed at the 30-minute
|
|
280
|
+
* rate.
|
|
281
|
+
*/
|
|
282
|
+
cacheWrite12hTokens: number;
|
|
283
|
+
}
|
|
284
|
+
/**
|
|
285
|
+
* Cost one OpenAI request.
|
|
286
|
+
*
|
|
287
|
+
* @beta
|
|
288
|
+
*/
|
|
289
|
+
export declare function openaiTokenCost(model: OpenAIModelId, usage: OpenAIUsageRecord): TokenCost;
|
|
226
290
|
//# sourceMappingURL=token-cost.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"token-cost.d.ts","sourceRoot":"","sources":["../../../src/utils/token-cost.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;
|
|
1
|
+
{"version":3,"file":"token-cost.d.ts","sourceRoot":"","sources":["../../../src/utils/token-cost.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAkD/E;;;;;;GAMG;AACH,MAAM,WAAW,UAAU;IACzB,gBAAgB,EAAE,MAAM,CAAC;IACzB,mBAAmB,EAAE,MAAM,CAAC;CAC7B;AAED;;;;GAIG;AACH,MAAM,WAAW,SAAS;IACxB,gFAAgF;IAChF,OAAO,EAAE,MAAM,CAAC;IAChB;;;;;;;;OAQG;IACH,QAAQ,EAAE,MAAM,CAAC;IACjB;;;;;;;OAOG;IACH,SAAS,EAAE,kBAAkB,CAAC;CAC/B;AAED;;;;GAIG;AACH,MAAM,WAAW,kBAAkB;IACjC,4DAA4D;IAC5D,SAAS,EAAE,MAAM,CAAC;IAClB,sDAAsD;IACtD,YAAY,EAAE,MAAM,CAAC;IACrB;;;;;;OAMG;IACH,aAAa,EAAE,MAAM,CAAC;IACtB,yEAAyE;IACzE,YAAY,EAAE,MAAM,CAAC;CACtB;AAiCD;;;;;;;;;;;;;GAaG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,gBAAgB,GAAG,UAAU,CAErE;AAED;;;;;;;;GAQG;AACH,eAAO,MAAM,+BAA+B,MAAM,CAAC;AACnD,YAAY;AACZ,eAAO,MAAM,mCAAmC,OAAO,CAAC;AACxD,YAAY;AACZ,eAAO,MAAM,mCAAmC,IAAI,CAAC;AAErD;;;;;;;;;GASG;AACH,MAAM,WAAW,oBAAoB;IACnC,wFAAwF;IACxF,mBAAmB,EAAE,MAAM,CAAC;IAC5B,4FAA4F;IAC5F,YAAY,EAAE,MAAM,CAAC;IACrB,uCAAuC;IACvC,eAAe,EAAE,MAAM,CAAC;IACxB,wEAAwE;IACxE,gBAAgB,EAAE,MAAM,CAAC;IACzB;;;;;;;;;OASG;IACH,kBAAkB,EAAE,MAAM,CAAC;CAC5B;AAED;;;;GAIG;AACH,wBAAgB,kBAAkB,CAChC,KAAK,EAAE,gBAAgB,EACvB,KAAK,EAAE,oBAAoB,GAC1B,SAAS,CAiCX;AAMD;;;;;;GAMG;AACH,eAAO,MAAM,6BAA6B,SAAU,CAAC;AAErD;;;;;;;;;GASG;AACH,eAAO,MAAM,8BAA8B,MAAM,CAAC;AAElD;;;;;;;;;;;GAWG;AACH,eAAO,MAAM,gCAAgC,QAAqC,CAAC;AAEnF;;;;;;;;GAQG;AACH,eAAO,MAAM,+BAA+B,QAAqC,CAAC;AAoElF;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAgB,cAAc,CAC5B,KAAK,EAAE,aAAa,EACpB,YAAY,EAAE,MAAM,EACpB,EAAE,GAAE,IAAiB,GACpB,UAAU,CASZ;AAED;;;;;;;;;;GAUG;AACH,MAAM,WAAW,iBAAiB;IAChC,kFAAkF;IAClF,YAAY,EAAE,MAAM,CAAC;IACrB,4CAA4C;IAC5C,eAAe,EAAE,MAAM,CAAC;IACxB;;;OAGG;IACH,aAAa,EAAE,MAAM,CAAC;IACtB,+EAA+E;IAC/E,YAAY,EAAE,MAAM,CAAC;CACtB;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,eAAe,CAC7B,KAAK,EAAE,aAAa,EACpB,KAAK,EAAE,iBAAiB,EACxB,EAAE,GAAE,IAAiB,GACpB,SAAS,CAuBX;AAuBD;;;;;;;GAOG;AACH,eAAO,MAAM,6BAA6B,SAAU,CAAC;AACrD,YAAY;AACZ,eAAO,MAAM,oCAAoC,IAAI,CAAC;AACtD,YAAY;AACZ,eAAO,MAAM,qCAAqC,MAAM,CAAC;AAEzD;;;;;;GAMG;AACH,eAAO,MAAM,iCAAiC,OAAO,CAAC;AACtD,YAAY;AACZ,eAAO,MAAM,iCAAiC,MAAM,CAAC;AAErD;;;;;GAKG;AACH,wBAAgB,cAAc,CAC5B,KAAK,EAAE,aAAa,EACpB,YAAY,EAAE,MAAM,GACnB,UAAU,GAAG;IAAE,gBAAgB,EAAE,MAAM,CAAA;CAAE,CAU3C;AAED;;;;;;;;;GASG;AACH,MAAM,WAAW,iBAAiB;IAChC,iFAAiF;IACjF,WAAW,EAAE,MAAM,CAAC;IACpB,8EAA8E;IAC9E,YAAY,EAAE,MAAM,CAAC;IACrB,gEAAgE;IAChE,eAAe,EAAE,MAAM,CAAC;IACxB,oFAAoF;IACpF,gBAAgB,EAAE,MAAM,CAAC;IACzB;;;;OAIG;IACH,mBAAmB,EAAE,MAAM,CAAC;CAC7B;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,aAAa,EAAE,KAAK,EAAE,iBAAiB,GAAG,SAAS,CAiCzF"}
|
|
@@ -76,5 +76,43 @@ export declare function enforceAnthropicToolSchema(parameters: SchemaNode, toolN
|
|
|
76
76
|
* than us silently picking a branch.
|
|
77
77
|
*/
|
|
78
78
|
export declare function toGeminiSchema(parameters: SchemaNode): SchemaNode;
|
|
79
|
+
/** The outcome of adapting a schema for OpenAI strict mode. */
|
|
80
|
+
export interface OpenAIStrictSchema {
|
|
81
|
+
/** The adapted schema — safe to send with `strict: true`. Always a fresh object. */
|
|
82
|
+
schema: SchemaNode;
|
|
83
|
+
/**
|
|
84
|
+
* JSON Pointers of properties that were OPTIONAL and are now required-but-nullable. Strict mode
|
|
85
|
+
* has no optional properties: the model must emit every key, sending `null` for an absent value.
|
|
86
|
+
* A tool handler that branches on `key in args` must branch on `args[key] == null` instead.
|
|
87
|
+
*/
|
|
88
|
+
madeNullable: string[];
|
|
89
|
+
/** JSON Pointers of `additionalProperties` values other than `false` that were closed. */
|
|
90
|
+
overridden: string[];
|
|
91
|
+
/** JSON Pointers of keywords strict mode cannot express, left in place. */
|
|
92
|
+
undroppable: string[];
|
|
93
|
+
}
|
|
94
|
+
/**
|
|
95
|
+
* Adapt a JSON Schema for OpenAI strict mode (`strict: true` on a function tool, or a
|
|
96
|
+
* `text.format` json_schema).
|
|
97
|
+
*
|
|
98
|
+
* Strict mode needs three things the canonical dialect does not:
|
|
99
|
+
*
|
|
100
|
+
* 1. **Every property listed in `required`.** A property the caller left optional is made
|
|
101
|
+
* required-but-nullable, which is the documented way to express "may be absent" — so the
|
|
102
|
+
* meaning survives, but the SHAPE of the arguments changes (absent becomes `null`), and that is
|
|
103
|
+
* reported in {@link OpenAIStrictSchema.madeNullable}.
|
|
104
|
+
* 2. **`additionalProperties: false` on every object**, as with Anthropic enforcement.
|
|
105
|
+
* 3. **No `allOf` / `not` / conditionals**, which are left in place so the request is rejected
|
|
106
|
+
* rather than silently enforcing a weaker schema.
|
|
107
|
+
*
|
|
108
|
+
* Numeric bounds, `pattern`, `format`, `minItems`/`maxItems`, `anyOf`, `$defs`/`$ref` are all
|
|
109
|
+
* supported by strict mode and pass through untouched.
|
|
110
|
+
*/
|
|
111
|
+
export declare function toOpenAIStrictSchema(parameters: SchemaNode): OpenAIStrictSchema;
|
|
112
|
+
/**
|
|
113
|
+
* {@link toOpenAIStrictSchema} for transport use: logs what changed and returns only the schema. A
|
|
114
|
+
* verb, for the same reason as {@link enforceAnthropicToolSchema}.
|
|
115
|
+
*/
|
|
116
|
+
export declare function enforceOpenAIToolSchema(parameters: SchemaNode, toolName: string): SchemaNode;
|
|
79
117
|
export {};
|
|
80
118
|
//# sourceMappingURL=tool-schema.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"tool-schema.d.ts","sourceRoot":"","sources":["../../../src/utils/tool-schema.ts"],"names":[],"mappings":"AAEA;;;GAGG;AACH,KAAK,UAAU,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAwH1C,wDAAwD;AACxD,MAAM,WAAW,cAAc;IAC7B,gEAAgE;IAChE,MAAM,EAAE,UAAU,CAAC;IACnB,iFAAiF;IACjF,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB;;;;;OAKG;IACH,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,oFAAoF;IACpF,WAAW,EAAE,MAAM,EAAE,CAAC;CACvB;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,wBAAgB,yBAAyB,CAAC,UAAU,EAAE,UAAU,GAAG,cAAc,CA6ChF;AAED;;;;;;;;;GASG;AACH,wBAAgB,0BAA0B,CAAC,UAAU,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,GAAG,UAAU,CAuB/F;AA0ED;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAgB,cAAc,CAAC,UAAU,EAAE,UAAU,GAAG,UAAU,CAsDjE"}
|
|
1
|
+
{"version":3,"file":"tool-schema.d.ts","sourceRoot":"","sources":["../../../src/utils/tool-schema.ts"],"names":[],"mappings":"AAEA;;;GAGG;AACH,KAAK,UAAU,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAwH1C,wDAAwD;AACxD,MAAM,WAAW,cAAc;IAC7B,gEAAgE;IAChE,MAAM,EAAE,UAAU,CAAC;IACnB,iFAAiF;IACjF,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB;;;;;OAKG;IACH,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,oFAAoF;IACpF,WAAW,EAAE,MAAM,EAAE,CAAC;CACvB;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,wBAAgB,yBAAyB,CAAC,UAAU,EAAE,UAAU,GAAG,cAAc,CA6ChF;AAED;;;;;;;;;GASG;AACH,wBAAgB,0BAA0B,CAAC,UAAU,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,GAAG,UAAU,CAuB/F;AA0ED;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAgB,cAAc,CAAC,UAAU,EAAE,UAAU,GAAG,UAAU,CAsDjE;AAaD,+DAA+D;AAC/D,MAAM,WAAW,kBAAkB;IACjC,oFAAoF;IACpF,MAAM,EAAE,UAAU,CAAC;IACnB;;;;OAIG;IACH,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,0FAA0F;IAC1F,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,2EAA2E;IAC3E,WAAW,EAAE,MAAM,EAAE,CAAC;CACvB;AAwBD;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,oBAAoB,CAAC,UAAU,EAAE,UAAU,GAAG,kBAAkB,CAoC/E;AAED;;;GAGG;AACH,wBAAgB,uBAAuB,CAAC,UAAU,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,GAAG,UAAU,CAsB5F"}
|
package/dist/esm/ai-provider.js
CHANGED
|
@@ -7,8 +7,8 @@ import { OpenAIProvider } from './providers/openai-provider';
|
|
|
7
7
|
import { AnthropicTransport } from './transports/anthropic-transport';
|
|
8
8
|
import { ChromePromptTransport } from './transports/chrome-prompt-transport';
|
|
9
9
|
import { GeminiTransport } from './transports/gemini-transport';
|
|
10
|
-
import {
|
|
11
|
-
import { SUPPORTED_ANTHROPIC_MODEL_IDS, SUPPORTED_GEMINI_MODEL_IDS, } from './types';
|
|
10
|
+
import { OpenAITransport } from './transports/openai-transport';
|
|
11
|
+
import { SUPPORTED_ANTHROPIC_MODEL_IDS, SUPPORTED_GEMINI_MODEL_IDS, SUPPORTED_OPENAI_MODEL_IDS, } from './types';
|
|
12
12
|
import { logger } from './utils';
|
|
13
13
|
import { isAIFeatureEnabled } from './utils/feature-flags';
|
|
14
14
|
function isChromeConfig(config) {
|
|
@@ -20,13 +20,13 @@ function isGeminiConfig(config) {
|
|
|
20
20
|
function isAnthropicConfig(config) {
|
|
21
21
|
return 'providerType' in config && config.providerType === 'anthropic';
|
|
22
22
|
}
|
|
23
|
-
function
|
|
24
|
-
return
|
|
23
|
+
function isOpenAIConfig(config) {
|
|
24
|
+
return 'providerType' in config && config.providerType === 'openai';
|
|
25
25
|
}
|
|
26
26
|
/**
|
|
27
27
|
* Creates an AIProvider from config.
|
|
28
28
|
* Uses ChromeProvider when providerType is 'chrome'.
|
|
29
|
-
* Uses
|
|
29
|
+
* Uses GeminiProvider, AnthropicProvider or OpenAIProvider for their `providerType`.
|
|
30
30
|
* Otherwise returns DefaultAIProvider.
|
|
31
31
|
* When the AI feature flag is disabled (?feature.ai), always returns DefaultAIProvider.
|
|
32
32
|
*
|
|
@@ -64,15 +64,16 @@ export function createAIProvider(config) {
|
|
|
64
64
|
});
|
|
65
65
|
return new AnthropicProvider({ criteriaInstructions: config.criteriaInstructions }, transport);
|
|
66
66
|
}
|
|
67
|
-
if (config &&
|
|
68
|
-
|
|
69
|
-
|
|
70
|
-
return new DefaultAIProvider();
|
|
71
|
-
}
|
|
72
|
-
logger.debug('createAIProvider: using OpenAIProvider (server proxy)');
|
|
73
|
-
const transport = new ServerOpenAITransport({
|
|
67
|
+
if (config && isOpenAIConfig(config)) {
|
|
68
|
+
logger.debug('createAIProvider: using OpenAIProvider (Responses)');
|
|
69
|
+
const transport = new OpenAITransport({
|
|
74
70
|
model: config.model,
|
|
75
71
|
timeout: config.timeout,
|
|
72
|
+
stallTimeout: config.stallTimeout,
|
|
73
|
+
serverEndpoint: config.serverEndpoint,
|
|
74
|
+
apiKey: config.apiKey,
|
|
75
|
+
maxTokens: config.maxTokens,
|
|
76
|
+
reasoningSummary: config.reasoningSummary,
|
|
76
77
|
});
|
|
77
78
|
return new OpenAIProvider({ criteriaInstructions: config.criteriaInstructions }, transport);
|
|
78
79
|
}
|
|
@@ -81,7 +82,9 @@ export function createAIProvider(config) {
|
|
|
81
82
|
}
|
|
82
83
|
/**
|
|
83
84
|
* Resolves AI config, optionally preferring Chrome when available.
|
|
84
|
-
* Use before registering AIProvider.
|
|
85
|
+
* Use before registering AIProvider. The cloud configs it returns carry no endpoint: add
|
|
86
|
+
* `serverEndpoint` (a server proxy) or `apiKey` (direct, trusted contexts only) before
|
|
87
|
+
* `createAIProvider`.
|
|
85
88
|
* Returns null when the AI feature flag is disabled (?feature.ai).
|
|
86
89
|
*
|
|
87
90
|
* @beta
|
|
@@ -117,7 +120,11 @@ export function resolveAIConfig() {
|
|
|
117
120
|
}
|
|
118
121
|
return { providerType: 'anthropic', model: anthropicModel };
|
|
119
122
|
}
|
|
120
|
-
|
|
123
|
+
const openaiModel = model !== null && model !== void 0 ? model : 'gpt-6-luna';
|
|
124
|
+
if (!SUPPORTED_OPENAI_MODEL_IDS.includes(openaiModel)) {
|
|
125
|
+
throw new Error(`resolveAIConfig: unsupported OpenAI model "${openaiModel}". Use one of: ${SUPPORTED_OPENAI_MODEL_IDS.join(', ')}.`);
|
|
126
|
+
}
|
|
127
|
+
return { providerType: 'openai', model: openaiModel };
|
|
121
128
|
});
|
|
122
129
|
}
|
|
123
130
|
/**
|
package/dist/esm/index.js
CHANGED
|
@@ -9,6 +9,8 @@ export { GeminiProvider } from './providers/gemini-provider';
|
|
|
9
9
|
export { AnthropicTransport, ResponseTruncatedError } from './transports/anthropic-transport';
|
|
10
10
|
export { BUDGETED_VENDORS, BudgetExhaustedError, DEFAULT_BUDGET_EXHAUSTED_MESSAGE, isBudgetedVendor, VENDOR_LABELS, vendorTypeOfLabel, } from './transports/budget-exhausted-error';
|
|
11
11
|
export { GeminiTransport, MalformedFunctionCallError } from './transports/gemini-transport';
|
|
12
|
+
export { OpenAITransport } from './transports/openai-transport';
|
|
13
|
+
export { OpenAIProvider } from './providers/openai-provider';
|
|
12
14
|
// The PROVIDER wall (GENC-1506) — the upstream vendor refusing this account, as distinct from our own
|
|
13
15
|
// spend budget above. The message constant is exported so ai-service's generation surfaces send the
|
|
14
16
|
// SAME sentence as the browser driver: the copy is provisional and a second copy would drift on the
|
|
@@ -34,10 +36,10 @@ export { ChatTemperature } from './utils/temperature';
|
|
|
34
36
|
// provider usage rather than `ChatMessage`, because Anthropic's per-TTL cache-write split exists
|
|
35
37
|
// only in the provider response and the path that needs to compute a cost is the one where no
|
|
36
38
|
// transport stamped one. See the module header for why re-pricing a stamped message is wrong.
|
|
37
|
-
export { ANTHROPIC_CACHE_READ_MULTIPLIER, ANTHROPIC_CACHE_WRITE_1H_MULTIPLIER, ANTHROPIC_CACHE_WRITE_5M_MULTIPLIER, anthropicRatesFor, anthropicTokenCost, GEMINI_CACHED_INPUT_MULTIPLIER, GEMINI_LONG_CONTEXT_THRESHOLD, geminiRatesFor, geminiTokenCost, } from './utils/token-cost';
|
|
39
|
+
export { ANTHROPIC_CACHE_READ_MULTIPLIER, ANTHROPIC_CACHE_WRITE_1H_MULTIPLIER, ANTHROPIC_CACHE_WRITE_5M_MULTIPLIER, anthropicRatesFor, anthropicTokenCost, GEMINI_CACHED_INPUT_MULTIPLIER, GEMINI_LONG_CONTEXT_THRESHOLD, geminiRatesFor, geminiTokenCost, OPENAI_CACHE_WRITE_12H_MULTIPLIER, OPENAI_CACHE_WRITE_30M_MULTIPLIER, OPENAI_LONG_CONTEXT_INPUT_MULTIPLIER, OPENAI_LONG_CONTEXT_OUTPUT_MULTIPLIER, OPENAI_LONG_CONTEXT_THRESHOLD, openaiRatesFor, openaiTokenCost, } from './utils/token-cost';
|
|
38
40
|
// `vendorOfModel` sits with the allowlists it is derived from, so it cannot drift from them.
|
|
39
41
|
// Exported for the pricing above: a caller holding a model id needs it to choose between the
|
|
40
42
|
// two vendor cost functions, and without it writes its own model→vendor map.
|
|
41
|
-
export { SUPPORTED_ANTHROPIC_MODEL_IDS, SUPPORTED_GEMINI_MODEL_IDS, vendorOfModel } from './types';
|
|
43
|
+
export { SUPPORTED_ANTHROPIC_MODEL_IDS, SUPPORTED_GEMINI_MODEL_IDS, SUPPORTED_OPENAI_MODEL_IDS, vendorOfModel, } from './types';
|
|
42
44
|
export { isChatToolCallUnknown } from './types/chat.types';
|
|
43
45
|
export { isSpeechRecognitionAvailable, startSpeechRecognition } from './utils/speech-to-text';
|
|
@@ -1,13 +1,9 @@
|
|
|
1
1
|
import { __awaiter } from "tslib";
|
|
2
2
|
import { buildCriteriaSystemPrompt, buildCriteriaUserPrompt, CRITERIA_JSON_SCHEMA, parseCriteriaResponse, } from '../interactions';
|
|
3
|
-
import { ServerOpenAITransport } from '../transports/server-openai-transport';
|
|
4
3
|
import { logger } from '../utils';
|
|
5
4
|
/**
|
|
6
|
-
* General OpenAI provider
|
|
7
|
-
*
|
|
8
|
-
*
|
|
9
|
-
* @remarks
|
|
10
|
-
* Configure via createAIProvider with ServerAIConfig (provider, model).
|
|
5
|
+
* General OpenAI provider, on `OpenAITransport` (the Responses API, direct or via a server proxy):
|
|
6
|
+
* criteria, prompt and chat — the same shape as its Gemini and Anthropic twins.
|
|
11
7
|
*
|
|
12
8
|
* @beta
|
|
13
9
|
*/
|
|
@@ -18,9 +14,7 @@ export class OpenAIProvider {
|
|
|
18
14
|
}
|
|
19
15
|
getStatus() {
|
|
20
16
|
return __awaiter(this, void 0, void 0, function* () {
|
|
21
|
-
const config = this.transport
|
|
22
|
-
? this.transport.getConfig()
|
|
23
|
-
: { provider: 'openai', model: 'gpt-4o-mini', contextLimit: undefined };
|
|
17
|
+
const config = this.transport.getConfig();
|
|
24
18
|
return { provider: 'openai', model: config.model, contextLimit: config.contextLimit };
|
|
25
19
|
});
|
|
26
20
|
}
|
|
@@ -40,7 +34,19 @@ export class OpenAIProvider {
|
|
|
40
34
|
}
|
|
41
35
|
catch (e) {
|
|
42
36
|
logger.error('OpenAIProvider interpretCriteria failed:', e);
|
|
37
|
+
return null;
|
|
43
38
|
}
|
|
44
39
|
});
|
|
45
40
|
}
|
|
41
|
+
prompt(message, options) {
|
|
42
|
+
return __awaiter(this, void 0, void 0, function* () {
|
|
43
|
+
const response = yield this.transport.sendChatMessage([], message, options);
|
|
44
|
+
return response.content;
|
|
45
|
+
});
|
|
46
|
+
}
|
|
47
|
+
chat(history, userMessage, options) {
|
|
48
|
+
return __awaiter(this, void 0, void 0, function* () {
|
|
49
|
+
return this.transport.sendChatMessage(history, userMessage, options);
|
|
50
|
+
});
|
|
51
|
+
}
|
|
46
52
|
}
|
|
@@ -60,19 +60,18 @@ export const VENDOR_LABELS = Object.freeze({
|
|
|
60
60
|
* and if it is reachable, `blocked` then derives true and locks a composer with
|
|
61
61
|
* headroom left.
|
|
62
62
|
*
|
|
63
|
-
*
|
|
64
|
-
*
|
|
65
|
-
*
|
|
66
|
-
*
|
|
67
|
-
*
|
|
68
|
-
* Anthropic and Gemini alone. So it can never raise a 402, and a 402 is never
|
|
69
|
-
* a statement about it.
|
|
63
|
+
* `chrome` is excluded for that reason: it runs on-device — no pot to exhaust, and free.
|
|
64
|
+
*
|
|
65
|
+
* `openai` is metered (GENC-1829): ai-service budgets it on its own `SpentUsdOpenai` counter and
|
|
66
|
+
* proxies it on `/api/openai/chat`. The proxy computes `otherVendorAvailable` over the vendors it
|
|
67
|
+
* OFFERS, so with OpenAI switched off there a 402 still says nothing about it.
|
|
70
68
|
*
|
|
71
69
|
* @beta
|
|
72
70
|
*/
|
|
73
71
|
export const BUDGETED_VENDORS = Object.freeze([
|
|
74
72
|
'anthropic',
|
|
75
73
|
'gemini',
|
|
74
|
+
'openai',
|
|
76
75
|
]);
|
|
77
76
|
/**
|
|
78
77
|
* Whether this vendor's spend is metered by the proxy — see
|
|
@@ -51,6 +51,8 @@ const CONTEXT_OVERFLOW_PHRASES = Object.freeze([
|
|
|
51
51
|
/input token count .* exceeds/i,
|
|
52
52
|
/maximum context length/i,
|
|
53
53
|
/context[_ ]length[_ ]exceeded/i,
|
|
54
|
+
// OpenAI Responses: `Your input exceeds the context window of this model.`
|
|
55
|
+
/exceeds the context window/i,
|
|
54
56
|
/too many (input )?tokens/i,
|
|
55
57
|
]);
|
|
56
58
|
/**
|