@jaypie/llm 1.3.9 → 1.3.11

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 (42) hide show
  1. package/dist/cjs/Llm.d.ts +19 -11
  2. package/dist/cjs/constants.d.ts +53 -4
  3. package/dist/cjs/errors/LlmError.d.ts +62 -0
  4. package/dist/cjs/errors/toLlmError.d.ts +11 -0
  5. package/dist/cjs/index.cjs +739 -177
  6. package/dist/cjs/index.cjs.map +1 -1
  7. package/dist/cjs/index.d.cts +177 -19
  8. package/dist/cjs/index.d.ts +5 -1
  9. package/dist/cjs/operate/adapters/AnthropicAdapter.d.ts +3 -2
  10. package/dist/cjs/operate/adapters/BedrockAdapter.d.ts +1 -1
  11. package/dist/cjs/operate/adapters/GoogleAdapter.d.ts +1 -1
  12. package/dist/cjs/operate/adapters/OpenAiAdapter.d.ts +7 -1
  13. package/dist/cjs/operate/adapters/OpenRouterAdapter.d.ts +5 -1
  14. package/dist/cjs/operate/adapters/XaiAdapter.d.ts +11 -1
  15. package/dist/cjs/operate/retry/RetryExecutor.d.ts +13 -0
  16. package/dist/cjs/operate/types.d.ts +15 -1
  17. package/dist/cjs/providers/google/types.d.ts +5 -0
  18. package/dist/cjs/types/LlmProvider.interface.d.ts +18 -0
  19. package/dist/cjs/util/classifyProviderError.d.ts +13 -0
  20. package/dist/cjs/util/effort.d.ts +42 -0
  21. package/dist/cjs/util/resolveModelChain.d.ts +17 -0
  22. package/dist/esm/Llm.d.ts +19 -11
  23. package/dist/esm/constants.d.ts +53 -4
  24. package/dist/esm/errors/LlmError.d.ts +62 -0
  25. package/dist/esm/errors/toLlmError.d.ts +11 -0
  26. package/dist/esm/index.d.ts +177 -19
  27. package/dist/esm/index.js +705 -148
  28. package/dist/esm/index.js.map +1 -1
  29. package/dist/esm/operate/adapters/AnthropicAdapter.d.ts +3 -2
  30. package/dist/esm/operate/adapters/BedrockAdapter.d.ts +1 -1
  31. package/dist/esm/operate/adapters/GoogleAdapter.d.ts +1 -1
  32. package/dist/esm/operate/adapters/OpenAiAdapter.d.ts +7 -1
  33. package/dist/esm/operate/adapters/OpenRouterAdapter.d.ts +5 -1
  34. package/dist/esm/operate/adapters/XaiAdapter.d.ts +11 -1
  35. package/dist/esm/operate/retry/RetryExecutor.d.ts +13 -0
  36. package/dist/esm/operate/types.d.ts +15 -1
  37. package/dist/esm/providers/google/types.d.ts +5 -0
  38. package/dist/esm/types/LlmProvider.interface.d.ts +18 -0
  39. package/dist/esm/util/classifyProviderError.d.ts +13 -0
  40. package/dist/esm/util/effort.d.ts +42 -0
  41. package/dist/esm/util/resolveModelChain.d.ts +17 -0
  42. package/package.json +1 -1
@@ -0,0 +1,42 @@
1
+ import { type LlmEffort } from "../constants.js";
2
+ /**
3
+ * Result of translating the provider-neutral {@link LlmEffort} scale to a
4
+ * provider's native reasoning control.
5
+ */
6
+ export interface LlmEffortMapping<T extends string | number = string> {
7
+ /**
8
+ * True when the requested neutral level had no distinct native rung and was
9
+ * collapsed or clamped onto a neighbor (e.g. `highest` -> Grok `high`, or
10
+ * `highest` -> OpenAI `high` on a model predating `xhigh`). Adapters log
11
+ * these at debug so a papered-over request stays on the record.
12
+ */
13
+ papered: boolean;
14
+ /** Native effort value for the target provider. */
15
+ value: T;
16
+ }
17
+ /**
18
+ * Per-provider translation of the provider-neutral {@link LlmEffort} scale.
19
+ *
20
+ * The neutral enum (lowest → low → medium → high → highest) is a relative
21
+ * five-point scale. Each table below maps it onto the target provider's native
22
+ * effort control, keeping `medium`/`high` semantically aligned across providers
23
+ * and using each provider's extra bottom (`minimal`) or top (`xhigh`/`max`)
24
+ * rung where one exists. Providers with fewer levels collapse the ends and mark
25
+ * the result `papered`. A single `effort` value is therefore safe to reuse
26
+ * across providers and fallback chains.
27
+ */
28
+ /** Consistent debug message for a papered-over effort level. */
29
+ export declare function paperedEffortMessage({ model, provider, requested, value, }: {
30
+ model: string;
31
+ provider: string;
32
+ requested: LlmEffort;
33
+ value: string | number;
34
+ }): string;
35
+ export declare function toOpenAiEffort(effort: LlmEffort, { model }: {
36
+ model: string;
37
+ }): LlmEffortMapping;
38
+ export declare function toXaiEffort(effort: LlmEffort): LlmEffortMapping;
39
+ export declare function toAnthropicEffort(effort: LlmEffort): LlmEffortMapping;
40
+ export declare function toGeminiThinkingLevel(effort: LlmEffort): LlmEffortMapping;
41
+ export declare function toGeminiThinkingBudget(effort: LlmEffort): LlmEffortMapping<number>;
42
+ export declare function toOpenRouterEffort(effort: LlmEffort): LlmEffortMapping;
@@ -0,0 +1,17 @@
1
+ import { LlmFallbackConfig } from "../types/LlmProvider.interface.js";
2
+ /**
3
+ * Normalizes a `model` option that may be a single model name or a
4
+ * preference-ordered array of model names.
5
+ *
6
+ * A `string[]` is interpreted as a fallback chain: index `0` is the primary
7
+ * model, indices `1..` become fallback entries with an auto-detected provider
8
+ * (reusing `determineModelProvider`, exactly as the constructor's first
9
+ * argument does).
10
+ *
11
+ * Returns the primary model plus the derived fallback chain. A bare string
12
+ * yields an empty chain and is unchanged.
13
+ */
14
+ export declare function resolveModelChain(model?: string | string[]): {
15
+ model?: string;
16
+ fallback: LlmFallbackConfig[];
17
+ };
package/dist/esm/Llm.d.ts CHANGED
@@ -1,15 +1,19 @@
1
1
  import { JsonObject } from "@jaypie/types";
2
2
  import { LlmProviderName } from "./constants.js";
3
- import { LlmFallbackConfig, LlmHistory, LlmInputMessage, LlmMessageOptions, LlmOperateInput, LlmOperateOptions, LlmOperateResponse, LlmOptions, LlmProvider } from "./types/LlmProvider.interface.js";
3
+ import { LlmFallbackConfig, LlmHistory, LlmInputMessage, LlmMessageOptions, LlmModelOption, LlmOperateInput, LlmOperateOptions, LlmOperateResponse, LlmOptions, LlmProvider } from "./types/LlmProvider.interface.js";
4
4
  import { LlmStreamChunk } from "./types/LlmStreamChunk.interface.js";
5
5
  declare class Llm implements LlmProvider {
6
6
  private _fallbackConfig?;
7
7
  private _llm;
8
8
  private _options;
9
9
  private _provider;
10
- constructor(providerName?: LlmProviderName | string, options?: LlmOptions);
10
+ constructor(providerName?: LlmProviderName | string, options?: Omit<LlmOptions, "model"> & {
11
+ model?: LlmModelOption;
12
+ });
11
13
  private createProvider;
12
- send(message: string, options?: LlmMessageOptions): Promise<string | JsonObject>;
14
+ send(message: string, options?: Omit<LlmMessageOptions, "model"> & {
15
+ model?: LlmModelOption;
16
+ }): Promise<string | JsonObject>;
13
17
  /**
14
18
  * Resolves the fallback chain from instance config and per-call options.
15
19
  * Per-call options take precedence over instance config.
@@ -20,23 +24,27 @@ declare class Llm implements LlmProvider {
20
24
  * Creates a fallback Llm instance lazily when needed.
21
25
  */
22
26
  private createFallbackInstance;
23
- operate(input: string | LlmHistory | LlmInputMessage | LlmOperateInput, options?: LlmOperateOptions): Promise<LlmOperateResponse>;
24
- stream(input: string | LlmHistory | LlmInputMessage | LlmOperateInput, options?: LlmOperateOptions): AsyncIterable<LlmStreamChunk>;
25
- static send(message: string, options?: LlmMessageOptions & {
27
+ operate(input: string | LlmHistory | LlmInputMessage | LlmOperateInput, options?: Omit<LlmOperateOptions, "model"> & {
28
+ model?: LlmModelOption;
29
+ }): Promise<LlmOperateResponse>;
30
+ stream(input: string | LlmHistory | LlmInputMessage | LlmOperateInput, options?: Omit<LlmOperateOptions, "model"> & {
31
+ model?: LlmModelOption;
32
+ }): AsyncIterable<LlmStreamChunk>;
33
+ static send(message: string, options?: Omit<LlmMessageOptions, "model"> & {
26
34
  llm?: LlmProviderName;
27
35
  apiKey?: string;
28
- model?: string;
36
+ model?: string | string[];
29
37
  }): Promise<string | JsonObject>;
30
- static operate(input: string | LlmHistory | LlmInputMessage | LlmOperateInput, options?: LlmOperateOptions & {
38
+ static operate(input: string | LlmHistory | LlmInputMessage | LlmOperateInput, options?: Omit<LlmOperateOptions, "model"> & {
31
39
  apiKey?: string;
32
40
  fallback?: LlmFallbackConfig[] | false;
33
41
  llm?: LlmProviderName;
34
- model?: string;
42
+ model?: string | string[];
35
43
  }): Promise<LlmOperateResponse>;
36
- static stream(input: string | LlmHistory | LlmInputMessage | LlmOperateInput, options?: LlmOperateOptions & {
44
+ static stream(input: string | LlmHistory | LlmInputMessage | LlmOperateInput, options?: Omit<LlmOperateOptions, "model"> & {
37
45
  llm?: LlmProviderName;
38
46
  apiKey?: string;
39
- model?: string;
47
+ model?: string | string[];
40
48
  }): AsyncIterable<LlmStreamChunk>;
41
49
  }
42
50
  export default Llm;
@@ -1,3 +1,20 @@
1
+ /**
2
+ * Provider-neutral reasoning-effort levels — a five-point relative scale that
3
+ * deliberately borrows no provider's vocabulary. Each adapter translates these
4
+ * to its provider's native control (OpenAI `reasoning.effort`, Anthropic
5
+ * `output_config.effort`, Gemini `thinkingLevel`/`thinkingBudget`, Grok
6
+ * `reasoning_effort`, OpenRouter `reasoning.effort`), spreading the scale across
7
+ * the provider's available range. Omitting `effort` leaves the provider default
8
+ * untouched, so it is safe to set across a fallback chain.
9
+ */
10
+ export declare const EFFORT: {
11
+ readonly LOWEST: "lowest";
12
+ readonly LOW: "low";
13
+ readonly MEDIUM: "medium";
14
+ readonly HIGH: "high";
15
+ readonly HIGHEST: "highest";
16
+ };
17
+ export type LlmEffort = (typeof EFFORT)[keyof typeof EFFORT];
1
18
  export declare const MODEL: {
2
19
  FABLE: string;
3
20
  OPUS: string;
@@ -7,13 +24,26 @@ export declare const MODEL: {
7
24
  GEMINI_FLASH: string;
8
25
  GEMINI_FLASH_LITE: string;
9
26
  GEMINI_PRO: string;
27
+ SOL: string;
28
+ TERRA: string;
29
+ LUNA: string;
30
+ /** @deprecated use MODEL.SOL (gpt-5.6-sol) */
10
31
  GPT: string;
32
+ /** @deprecated use MODEL.TERRA (gpt-5.6-terra) */
11
33
  GPT_MINI: string;
34
+ /** @deprecated use MODEL.LUNA (gpt-5.6-luna) */
12
35
  GPT_NANO: string;
13
36
  GROK: string;
37
+ OPENROUTER: {
38
+ GLM: string;
39
+ LUNA: string;
40
+ SONNET: string;
41
+ };
14
42
  };
15
43
  export declare const PROVIDER: {
16
44
  readonly BEDROCK: {
45
+ readonly DEFAULT: "amazon.nova-pro-v1:0";
46
+ /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.BEDROCK.DEFAULT. */
17
47
  readonly MODEL: {
18
48
  readonly DEFAULT: "amazon.nova-lite-v1:0";
19
49
  readonly LARGE: "amazon.nova-pro-v1:0";
@@ -25,16 +55,18 @@ export declare const PROVIDER: {
25
55
  readonly REGION: "AWS_REGION";
26
56
  };
27
57
  readonly ANTHROPIC: {
58
+ readonly DEFAULT: string;
28
59
  readonly MAX_TOKENS: {
29
60
  readonly DEFAULT: 16384;
30
61
  };
62
+ /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.ANTHROPIC.DEFAULT, or pick a specific model from MODEL.*. */
31
63
  readonly MODEL: {
32
64
  readonly DEFAULT: "claude-sonnet-4-6";
33
65
  readonly LARGE: "claude-opus-4-8";
34
66
  readonly SMALL: "claude-sonnet-4-6";
35
67
  readonly TINY: "claude-haiku-4-5";
36
68
  };
37
- readonly MODEL_MATCH_WORDS: readonly ["anthropic", "claude", "haiku", "opus", "sonnet"];
69
+ readonly MODEL_MATCH_WORDS: readonly ["anthropic", "claude", "fable", "haiku", "mythos", "opus", "sonnet"];
38
70
  readonly NAME: "anthropic";
39
71
  readonly PROMPT: {
40
72
  readonly AI: "\n\nAssistant:";
@@ -51,6 +83,8 @@ export declare const PROVIDER: {
51
83
  };
52
84
  /** @deprecated Use PROVIDER.GOOGLE — "Google" is the provider; Gemini is the model family */
53
85
  readonly GEMINI: {
86
+ readonly DEFAULT: string;
87
+ /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.GOOGLE.DEFAULT, or pick a specific model from MODEL.*. */
54
88
  readonly MODEL: {
55
89
  readonly DEFAULT: "gemini-3.1-pro-preview";
56
90
  readonly LARGE: "gemini-3.1-pro-preview";
@@ -65,6 +99,8 @@ export declare const PROVIDER: {
65
99
  };
66
100
  };
67
101
  readonly GOOGLE: {
102
+ readonly DEFAULT: string;
103
+ /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.GOOGLE.DEFAULT, or pick a specific model from MODEL.*. */
68
104
  readonly MODEL: {
69
105
  readonly DEFAULT: "gemini-3.1-pro-preview";
70
106
  readonly LARGE: "gemini-3.1-pro-preview";
@@ -79,16 +115,20 @@ export declare const PROVIDER: {
79
115
  };
80
116
  };
81
117
  readonly OPENAI: {
118
+ readonly DEFAULT: string;
119
+ /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.OPENAI.DEFAULT, or pick a specific model from MODEL.*. */
82
120
  readonly MODEL: {
83
121
  readonly DEFAULT: "gpt-5.4";
84
122
  readonly LARGE: "gpt-5.5";
85
123
  readonly SMALL: "gpt-5.4-mini";
86
124
  readonly TINY: "gpt-5.4-nano";
87
125
  };
88
- readonly MODEL_MATCH_WORDS: readonly ["openai", "gpt", RegExp];
126
+ readonly MODEL_MATCH_WORDS: readonly ["gpt", "luna", "openai", "sol", "terra", RegExp];
89
127
  readonly NAME: "openai";
90
128
  };
91
129
  readonly OPENROUTER: {
130
+ readonly DEFAULT: string;
131
+ /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.OPENROUTER.DEFAULT, or pick a specific route from MODEL.OPENROUTER.*. */
92
132
  readonly MODEL: {
93
133
  readonly DEFAULT: "anthropic/claude-sonnet-4-6";
94
134
  readonly LARGE: "anthropic/claude-opus-4-8";
@@ -107,6 +147,8 @@ export declare const PROVIDER: {
107
147
  readonly XAI: {
108
148
  readonly API_KEY: "XAI_API_KEY";
109
149
  readonly BASE_URL: "https://api.x.ai/v1";
150
+ readonly DEFAULT: string;
151
+ /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.XAI.DEFAULT, or pick a specific model from MODEL.*. */
110
152
  readonly MODEL: {
111
153
  readonly DEFAULT: "grok-latest";
112
154
  readonly LARGE: "grok-latest";
@@ -119,6 +161,7 @@ export declare const PROVIDER: {
119
161
  };
120
162
  export type LlmProviderName = typeof PROVIDER.ANTHROPIC.NAME | typeof PROVIDER.BEDROCK.NAME | typeof PROVIDER.GOOGLE.NAME | typeof PROVIDER.OPENAI.NAME | typeof PROVIDER.OPENROUTER.NAME | typeof PROVIDER.XAI.NAME;
121
163
  export declare const DEFAULT: {
164
+ /** @deprecated Size tiers are retired in 2.0. Use DEFAULT.PROVIDER.DEFAULT, or pick a specific model from MODEL.*. */
122
165
  readonly MODEL: {
123
166
  readonly BASE: "gpt-5.4";
124
167
  readonly LARGE: "gpt-5.5";
@@ -126,19 +169,25 @@ export declare const DEFAULT: {
126
169
  readonly TINY: "gpt-5.4-nano";
127
170
  };
128
171
  readonly PROVIDER: {
172
+ readonly DEFAULT: string;
173
+ /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.OPENAI.DEFAULT, or pick a specific model from MODEL.*. */
129
174
  readonly MODEL: {
130
175
  readonly DEFAULT: "gpt-5.4";
131
176
  readonly LARGE: "gpt-5.5";
132
177
  readonly SMALL: "gpt-5.4-mini";
133
178
  readonly TINY: "gpt-5.4-nano";
134
179
  };
135
- readonly MODEL_MATCH_WORDS: readonly ["openai", "gpt", RegExp];
180
+ readonly MODEL_MATCH_WORDS: readonly ["gpt", "luna", "openai", "sol", "terra", RegExp];
136
181
  readonly NAME: "openai";
137
182
  };
138
183
  };
184
+ /**
185
+ * @deprecated Size-tier catalogs are retired in 2.0. Pick specific models from
186
+ * MODEL.* (grouped by provider via MODEL_MATCH_WORDS / determineModelProvider).
187
+ */
139
188
  export declare const ALL: {
140
189
  readonly BASE: readonly ["claude-sonnet-4-6", "gemini-3.1-pro-preview", "gpt-5.4", "grok-latest"];
141
- readonly COMBINED: readonly ("claude-sonnet-4-6" | "claude-opus-4-8" | "claude-haiku-4-5" | "gemini-3.1-pro-preview" | "gemini-3.5-flash" | "gemini-3.1-flash-lite" | "gpt-5.4" | "gpt-5.5" | "gpt-5.4-mini" | "gpt-5.4-nano" | "grok-latest" | "grok-4-1-fast-reasoning" | "grok-4-1-fast-non-reasoning")[];
190
+ readonly COMBINED: readonly ("claude-opus-4-8" | "claude-haiku-4-5" | "gemini-3.5-flash" | "gemini-3.1-flash-lite" | "gemini-3.1-pro-preview" | "gpt-5.5" | "gpt-5.4-mini" | "gpt-5.4-nano" | "grok-latest" | "claude-sonnet-4-6" | "gpt-5.4" | "grok-4-1-fast-reasoning" | "grok-4-1-fast-non-reasoning")[];
142
191
  readonly LARGE: readonly ["claude-opus-4-8", "gemini-3.1-pro-preview", "gpt-5.5", "grok-latest"];
143
192
  readonly SMALL: readonly ["claude-sonnet-4-6", "gemini-3.5-flash", "gpt-5.4-mini", "grok-4-1-fast-reasoning"];
144
193
  readonly TINY: readonly ["claude-haiku-4-5", "gemini-3.1-flash-lite", "gpt-5.4-nano", "grok-4-1-fast-non-reasoning"];
@@ -0,0 +1,62 @@
1
+ import { JaypieError } from "@jaypie/errors";
2
+ import { ErrorCategory } from "../operate/types.js";
3
+ export interface LlmErrorOptions {
4
+ /** Provider name that produced the error (e.g. "anthropic", "google") */
5
+ provider?: string;
6
+ /** Model in use when the error occurred */
7
+ model?: string;
8
+ /** Milliseconds to wait before retrying, when the provider suggests one */
9
+ retryAfterMs?: number;
10
+ /** The original provider error, preserved for inspection */
11
+ cause?: unknown;
12
+ }
13
+ /**
14
+ * Normalized, provider-agnostic LLM error. Thrown by the operate/stream retry
15
+ * layer when a request cannot be completed, so consumers can `catch` a stable
16
+ * type regardless of which provider failed. Extends {@link JaypieError} so
17
+ * `isJaypieError()` and `.status` continue to work; the original provider error
18
+ * is preserved on `.cause`.
19
+ */
20
+ export declare class LlmError extends JaypieError {
21
+ readonly category: ErrorCategory;
22
+ readonly provider?: string;
23
+ readonly model?: string;
24
+ readonly retryAfterMs?: number;
25
+ readonly cause?: unknown;
26
+ constructor(message: string, category: ErrorCategory, { status, title, provider, model, retryAfterMs, cause, }?: LlmErrorOptions & {
27
+ status?: number;
28
+ title?: string;
29
+ });
30
+ }
31
+ /**
32
+ * Short-term rate limiting (per-minute 429). Terminal within the request
33
+ * budget; `retryAfterMs` carries the provider's suggested wait when available.
34
+ */
35
+ export declare class LlmRateLimitError extends LlmError {
36
+ constructor(message: string, options?: LlmErrorOptions);
37
+ }
38
+ /**
39
+ * Provider quota is exhausted or the account cannot be billed. Terminal and
40
+ * actionable. `reason` distinguishes an exhausted usage quota from insufficient
41
+ * funds.
42
+ */
43
+ export declare class LlmQuotaError extends LlmError {
44
+ readonly reason: "quota" | "billing";
45
+ constructor(message: string, { reason, ...options }?: LlmErrorOptions & {
46
+ reason?: "quota" | "billing";
47
+ });
48
+ }
49
+ /**
50
+ * The request cannot be recovered from (bad request, authentication,
51
+ * permission, not found). Terminal.
52
+ */
53
+ export declare class LlmUnrecoverableError extends LlmError {
54
+ constructor(message: string, options?: LlmErrorOptions);
55
+ }
56
+ /**
57
+ * A transient or unknown error survived the retry budget. Terminal only because
58
+ * retries were exhausted; the underlying condition may succeed later.
59
+ */
60
+ export declare class LlmTransientError extends LlmError {
61
+ constructor(message: string, options?: LlmErrorOptions);
62
+ }
@@ -0,0 +1,11 @@
1
+ import { ClassifiedError } from "../operate/types.js";
2
+ import { LlmError } from "./LlmError.js";
3
+ /**
4
+ * Map a {@link ClassifiedError} to the matching {@link LlmError} subclass,
5
+ * preserving the original error as `cause`. Used by the retry layer to throw a
6
+ * stable, catchable type instead of a bare gateway error.
7
+ */
8
+ export declare function toLlmError(classified: ClassifiedError, context?: {
9
+ provider?: string;
10
+ model?: string;
11
+ }): LlmError;