diff --git a/apps/memos-local-plugin/core/config/defaults.ts b/apps/memos-local-plugin/core/config/defaults.ts index 5c9dff305..188040070 100644 --- a/apps/memos-local-plugin/core/config/defaults.ts +++ b/apps/memos-local-plugin/core/config/defaults.ts @@ -44,6 +44,7 @@ export const DEFAULT_CONFIG: ResolvedConfig = { apiKey: "", timeoutMs: 45_000, maxRetries: 3, + maxTokens: 4_000, providerIgnore: [], providerOrder: [], openRouter: false, @@ -60,6 +61,7 @@ export const DEFAULT_CONFIG: ResolvedConfig = { apiKey: "", temperature: 0, timeoutMs: 60_000, + maxTokens: 4_000, providerIgnore: [], providerOrder: [], openRouter: false, @@ -74,6 +76,7 @@ export const DEFAULT_CONFIG: ResolvedConfig = { apiKey: "", temperature: 0, timeoutMs: 60_000, + maxTokens: 4_000, providerIgnore: [], providerOrder: [], openRouter: false, diff --git a/apps/memos-local-plugin/core/config/schema.ts b/apps/memos-local-plugin/core/config/schema.ts index 8566f90f3..0eec2a1df 100644 --- a/apps/memos-local-plugin/core/config/schema.ts +++ b/apps/memos-local-plugin/core/config/schema.ts @@ -91,6 +91,11 @@ const LlmSchema = Type.Object({ timeoutMs: NumberInRange(45_000, 1_000), /** Max retries on transient errors. */ maxRetries: NumberInRange(3, 0, 10), + /** + * Default output-token budget for LLM calls. A larger default prevents + * reasoning models from truncating structured JSON responses. + */ + maxTokens: NumberInRange(4_000, 1_024, 32_768), /** OpenRouter provider routing — providers to skip. */ providerIgnore: Type.Optional(Type.Array(Type.String(), { default: [] })), /** OpenRouter provider routing — preferred order. */ @@ -123,6 +128,11 @@ const SkillEvolverSchema = Type.Object({ apiKey: StringWithDefault(""), temperature: NumberInRange(0, 0, 2), timeoutMs: NumberInRange(60_000, 1_000), + /** + * Default output-token budget for dedicated skill-evolver and L3 calls. + * These calls commonly return larger structured JSON documents. + */ + maxTokens: NumberInRange(4_000, 1_024, 32_768), /** OpenRouter provider routing — providers to skip. */ providerIgnore: Type.Optional(Type.Array(Type.String(), { default: [] })), /** OpenRouter provider routing — preferred order. */ diff --git a/apps/memos-local-plugin/core/pipeline/memory-core.ts b/apps/memos-local-plugin/core/pipeline/memory-core.ts index a0354bf64..b14c2e132 100644 --- a/apps/memos-local-plugin/core/pipeline/memory-core.ts +++ b/apps/memos-local-plugin/core/pipeline/memory-core.ts @@ -133,6 +133,7 @@ type DedicatedLlmConfig = { apiKey?: string; temperature?: number; timeoutMs?: number; + maxTokens?: number; providerIgnore?: string[]; providerOrder?: string[]; openRouter?: boolean; @@ -433,6 +434,7 @@ export async function bootstrapMemoryCoreFull( apiKey: evolver?.apiKey ?? "", temperature: evolver?.temperature ?? 0, timeoutMs: evolver?.timeoutMs ?? 60_000, + maxTokens: evolver?.maxTokens, providerIgnore: evolver?.providerIgnore, providerOrder: evolver?.providerOrder, openRouter: evolver?.openRouter ?? false, @@ -495,6 +497,7 @@ export async function bootstrapMemoryCoreFull( apiKey: l3c?.apiKey ?? "", temperature: l3c?.temperature ?? 0, timeoutMs: l3c?.timeoutMs ?? 60_000, + maxTokens: l3c?.maxTokens, providerIgnore: l3c?.providerIgnore, providerOrder: l3c?.providerOrder, openRouter: l3c?.openRouter ?? false, diff --git a/apps/memos-local-plugin/tests/unit/config/load.test.ts b/apps/memos-local-plugin/tests/unit/config/load.test.ts index 5d85fd6e0..69b9d9478 100644 --- a/apps/memos-local-plugin/tests/unit/config/load.test.ts +++ b/apps/memos-local-plugin/tests/unit/config/load.test.ts @@ -56,6 +56,25 @@ describe("config/loadConfig", () => { expect(cfg.embedding.openRouter).toBe(false); }); + it("defaults and validates LLM output-token budgets", () => { + const defaults = resolveConfig({}); + expect(defaults.llm.maxTokens).toBe(4_000); + expect(defaults.skillEvolver.maxTokens).toBe(4_000); + expect(defaults.l3Llm.maxTokens).toBe(4_000); + + const configured = resolveConfig({ + llm: { maxTokens: 2_048 }, + skillEvolver: { maxTokens: 8_192 }, + l3Llm: { maxTokens: 16_384 }, + }); + expect(configured.llm.maxTokens).toBe(2_048); + expect(configured.skillEvolver.maxTokens).toBe(8_192); + expect(configured.l3Llm.maxTokens).toBe(16_384); + + expect(() => resolveConfig({ llm: { maxTokens: 1_023 } })).toThrow(/schema validation/); + expect(() => resolveConfig({ skillEvolver: { maxTokens: 32_769 } })).toThrow(/schema validation/); + }); + it("merges YAML over defaults and preserves unspecified branches", async () => { const yaml = ` viewer: diff --git a/apps/memos-local-plugin/tests/unit/llm/client.test.ts b/apps/memos-local-plugin/tests/unit/llm/client.test.ts index cd125ecd8..b12999e31 100644 --- a/apps/memos-local-plugin/tests/unit/llm/client.test.ts +++ b/apps/memos-local-plugin/tests/unit/llm/client.test.ts @@ -96,6 +96,17 @@ describe("llm/client", () => { expect(fake.lastMessages).toEqual([{ role: "user", content: "hi there" }]); }); + it("uses the configured maxTokens unless the call overrides it", async () => { + const fake = new FakeProvider("openai_compatible", () => ({ text: "ok", durationMs: 1 })); + const client = createLlmClientWithProvider(cfg({ maxTokens: 4_000 }), fake); + + await client.complete("configured budget"); + expect(fake.lastInput?.maxTokens).toBe(4_000); + + await client.complete("call override", { maxTokens: 6_000 }); + expect(fake.lastInput?.maxTokens).toBe(6_000); + }); + it("injects json hints into system and user messages when jsonMode=true", async () => { const fake = new FakeProvider("openai_compatible", () => ({ text: '{"ok":1}', durationMs: 1 })); const client = createLlmClientWithProvider(cfg(), fake); diff --git a/apps/memos-local-plugin/tests/unit/pipeline/bootstrap-llm-config.test.ts b/apps/memos-local-plugin/tests/unit/pipeline/bootstrap-llm-config.test.ts index 512c4cdde..71829cd9a 100644 --- a/apps/memos-local-plugin/tests/unit/pipeline/bootstrap-llm-config.test.ts +++ b/apps/memos-local-plugin/tests/unit/pipeline/bootstrap-llm-config.test.ts @@ -86,12 +86,14 @@ describe("bootstrapMemoryCore dedicated LLM config", () => { llm: provider: local_only model: main + maxTokens: 5000 skillEvolver: provider: openai_compatible endpoint: https://openrouter.ai/api/v1 openRouter: true model: skill-model apiKey: sk-test + maxTokens: 6000 providerIgnore: - together providerOrder: @@ -102,6 +104,7 @@ l3Llm: openRouter: true model: l3-model apiKey: sk-test + maxTokens: 7000 providerIgnore: - novita providerOrder: @@ -120,16 +123,21 @@ l3Llm: }); expect(capturedLlmConfigs.find((cfg) => cfg.model === "skill-model")).toMatchObject({ + maxTokens: 6_000, providerIgnore: ["together"], providerOrder: ["anthropic"], openRouter: true, }); expect(capturedLlmConfigs.find((cfg) => cfg.model === "l3-model")).toMatchObject({ + maxTokens: 7_000, providerIgnore: ["novita"], providerOrder: ["openai"], openRouter: true, reasoning: { enabled: true, maxTokens: 4_000 }, }); + expect(capturedLlmConfigs.find((cfg) => cfg.model === "main")).toMatchObject({ + maxTokens: 5_000, + }); }); it("normalizes missing dedicated OpenRouter flags to false", async () => {