0.2.0

What's new

Multi-provider support

Endpoints can now target Mistral and DeepSeek alongside OpenRouter — any OpenAI-compatible /models endpoint works.

Strict-UUID Integrations API

Endpoints pin to a specific integration row via integration_uuid. Legacy api_key values are auto-migrated with idempotency guards and rate-limited warnings.

Reasoning chain-of-thought

reasoning_effort and reasoning_max_tokens are captured in request history.

Security

SSRF guard on base_url blocks localhost, RFC1918, link-local, and IPv6 loopback/ULA addresses.

Bug fixes

• Provider-switch now fetches models from the new provider's base URL (was stale)
• Duplicated mask_api_key/1 consolidated into schema helper
• migrate_legacy/0 surfaces inner :error instead of masking as {:ok, _}
• Snapshot-based UUID lookups prevent N+1 queries during migration
• Integration deselect routes through changeset

Full changelog

https://hex.pm/packages/phoenix_kit_ai/0.2.0