v2.3.0

Added

• Offer the known LiteLLM proxy URL as the default during login (#150).

Fixed

• Make Moonshot reasoning suppression route-aware, preserving forced-thinking aliases and handling ambiguous duplicate routes conservatively (#152).

Documentation

• Document the Pi startup “No models available” discovery race and recovery steps (#149).

Thanks to @densilver for their first contribution.

Full changelog: v2.2.1...v2.3.0

v2.2.1

What's changed

• Added an explicit per-provider allowInsecureHttp opt-in for container-host, private-network, and other intentionally plaintext LiteLLM endpoints while keeping HTTPS and loopback defaults secure (#145).
• Applied the transport policy consistently across login, model discovery, MCP, and Skills, and documented the plaintext credential and request-data risk.

Thanks @kovjanos for reporting #144.

Full changelog: v2.2.0...v2.2.1

v2.2.0

Improved

• Populate LiteLLM models during startup again, so cold starts have a catalog without first opening /model (#142). Thanks to @mxr576 for reporting #137.
• Keep Skills prompt enrichment best-effort when credentials expire or the proxy is unavailable (#141).

Changed

• Stop contacting models.dev during fallback discovery; /v1/models now uses Pi catalog metadata and safe defaults only (#138).

Full changelog: v2.1.0...v2.2.0

v2.1.0

What’s new

• Added LiteLLM CLI SSO login support, including browser-based authentication, secure credential storage, API-key fallback, and safer handling of insecure endpoints and sensitive diagnostics (#134).

Full Changelog: v2.0.6...v2.1.0

v2.0.6

Reasoning and compatibility

• Map LiteLLM reasoning-effort capability flags into Pi thinking levels, including xhigh and max, while preserving explicit server capability overrides (#132). Thanks to @ihannsen for reporting #128.
• Support the Pi 0.84 model refresh context and update the development compatibility baseline (#129).

Security and release integrity

• Require HTTPS for remote LiteLLM endpoints outside loopback and redact credentials and remote response bodies from smoke and authentication errors (#131).
• Validate authorized signed release tags, pin smoke dependencies, block dependency install scripts, and remediate the nanoid advisory (#131).

Full changelog: v2.0.5...v2.0.6

v2.0.5

Model discovery

• Refresh stale cached catalog metadata while preserving partial or unknown entries (#124).
• Keep prototype-backed model stores working during refresh and persistence (#124).

Installation and security

• Keep Pi host peer packages optional so installing the extension does not install another Pi runtime (#125).
• Update the Pi development baseline and remediate vulnerable transitive dependencies (#125).

Full changelog: v2.0.4...v2.0.5

v2.0.4

Highlights

• Restored reasoning and native xhigh/max thinking levels for bare Opus 5 aliases exposed through LiteLLM.

Maintenance

• Updated the Pi development compatibility baseline to 0.83.0 while preserving Pi 0.81+ runtime compatibility.

Full changelog: v2.0.3...v2.0.4

v2.0.3

Highlights

• Improved Moonshot/Kimi tool-call compatibility, including preserving image attachments from tool results. Thanks to @vlad-deriv for contributing #113.
• Fixed Gemini/Vertex reasoning-effort casing through LiteLLM. Thanks to @josh-tustanowski-fdmt for reporting and contributing #120.
• Added correct reasoning metadata for bare Fable 5 aliases (#119).

Maintenance

• Updated development tooling dependencies (#116).

New contributors

• @vlad-deriv made their first contribution in #113.
• @josh-tustanowski-fdmt made their first contribution in #120.

Full changelog: v2.0.2...v2.0.3

v2.0.2

Highlights

• Fixed wildcard LiteLLM model entries so the concrete models returned by /v1/models are selectable (#107, #108). Special thanks to @indistinctTalk for reporting the issue and contributing the fix.
• Fixed /login litellm freezing while MCP discovery is in progress (#109, #111). Thanks to @A7exSchin for reporting the issue.
• Fixed Kimi/Moonshot-pattern OpenAI-compatible routes rejecting the injected thinking parameter (#110, #112). Thanks again to @A7exSchin for reporting it.

Maintenance

• Updated compatibility coverage to Pi 0.82.1 (#105).
• Updated PostCSS from 8.5.16 to 8.5.23 (#104).

Full changelog: v2.0.1...v2.0.2

v2.0.1

Fixed

• Pi-managed installations on Pi 0.81.1 and 0.82.0 can load the LiteLLM extension again. The provider now uses Pi's supported compatibility entry point instead of unresolvable deep module paths (#102).
• Resolves the extension-loading failures reported in #93 and #101.

Full Changelog: v2.0.0...v2.0.1