Provider presets and factory #44

Merged

approver_agent merged 5 commits from task/provider-presets_4d90a725-5a7c-4bee-bbe4-d036cfb69c5b into feature/5dd6e253-cc9b-41f3-9091-bb834f7a373a-llm-provider-layer 2026-10-04 09:18:35 -04:00

Conversation 1 Commits 5 Files Changed 8

+1055 -6

5 Commits

Author	SHA1	Message	Date
dev_agentandClaude Opus 5.5	08be8891da	Reject API keys that are not valid header values at construction ... FromSettings now fails with llm.ErrInvalidKey for a key with a control character such as CR or LF, for every preset that sends a key, and openai.New rejects such a key too, so neither defers the failure to the first request. anthropic.New already rejected it. No error quotes the key. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-04 09:08:46 -04:00
dev_agentandClaude Opus 5.5	31322f239e	Redact the API key from preset configuration and validation errors ... FromSettings errors and the configured provider's request-validation errors now pass through a presets redactedError: its text has the key removed with llm.Redact, it does not unwrap to the unredacted layers, and Is still matches the llm sentinels. A key that equals a request value (temperature -0.5) or a setting no longer appears in the error. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-04 09:03:12 -04:00
dev_agentandClaude Opus 5.5	cb93134f3e	Fix reasoning-model temperature quirk matching and validation ... - Omit temperature for OpenRouter's namespaced OpenAI reasoning models (openai/o1, openai/o1-mini, openai/o3-mini, openai/o1:online). - Match o1/o3 only as the bare family name or with a "-" variant, so o123 and o3custom keep their temperature. - Validate the caller's request before clearing temperature, so a negative temperature is still a validation error, not a sent request. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-04 08:58:43 -04:00
dev_agentandClaude Opus 5.5	b6e433cc98	Allow openai and openrouter AI providers in the schema ... 0001 limited ai_settings.provider and ai_runs.provider to openai_compatible, anthropic and ollama, so the openai and openrouter presets could not be stored. Migration 0004 rebuilds both tables with the wider CHECK, keeping the three existing values and every row. ai_runs is referenced by suggestion_batches and week_plans, so it is rebuilt by copying rows aside and recreating the table under its own name with foreign keys deferred; the create-then-rename rebuild from 0003 leaves the dropped table's references counted as violations at commit. Tests cover upgrading a database with ai_settings, ai_runs and referencing rows, and storing every preset provider in both tables. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-04 08:54:03 -04:00
dev_agentandClaude Opus 5.5	61c767ea32	Add provider presets and FromSettings factory ... internal/ai/llm/presets defines the anthropic, openai, openrouter, ollama and openai_compatible presets (base URL, adapter, key policy, headers, structured output modes and model quirks) and FromSettings, which turns llm.Settings plus the decrypted key into a ready llm.Provider or a typed configuration error from package llm. Temperature is omitted for model ids matching ^o1 or ^o3, listed once in TemperatureRejectingModels. Maestro task: 4d90a725-5a7c-4bee-bbe4-d036cfb69c5b Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-04 08:51:19 -04:00