- options.go: new WithChatCompletions(resolver) and
WithChatCompletionsPath(path); api.New(...) now auto-mounts at
/v1/chat/completions when a resolver is configured (previously the
resolver could be attached but never mounted, which would have
panicked Gin)
- chat_completions.go: fixed missing net/http import, dropped
ModelType during discovery, Retry-After header set after c.JSON
silently lost, swapped OpenAI error type/code fields, swapped
validate call site, redundant nil check, builder length read before
nil-receiver check
- openapi.go: effective*Path helpers surface an explicit path even
when the corresponding Enabled flag is false so CLI callers still
get x-*-path extensions; /swagger always in authentik public paths
- chat_completions_test.go: Good/Bad/Ugly coverage for new options,
validation, no-resolver behaviour
- openapi_test.go: fix stale assertion for CacheEnabled-gated X-Cache
- go.mod: bump dappco.re/go/core/cli to v0.5.2
- Removed local go-io / go-log stubs — replace points to outer
modules for single source of truth
- Migrated forge.lthn.ai/core/cli imports to dappco.re/go/core/cli
across cmd/api/*.go + docs
Co-Authored-By: Virgil <virgil@lethean.io>
Prevent silent unbounded cache creation when both maxEntries and maxBytes
are non-positive: newCacheStore now returns nil, WithCacheLimits skips
middleware registration, and WithCache defaults to 1 000-entry LRU cap
when called with only a TTL argument.
Co-Authored-By: Virgil <virgil@lethean.io>
Adds X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset to successful responses and 429 rejections, and documents the headers in OpenAPI.
Co-Authored-By: Virgil <virgil@lethean.io>
Adds a token-bucket WithRateLimit option that rejects excess requests with 429 Too Many Requests and a standard error envelope.
Co-Authored-By: Virgil <virgil@lethean.io>
