codex-kimi-bridge v0.3.0

codex-kimi-bridge 0.3.0

Highlights

• Translate Codex Responses namespace containers whose children are function or custom tools.
• Flatten namespaced tools to collision-safe Kimi Chat function names of at most 64 characters.
• Restore the original namespace + name in non-streaming responses, streaming events, and later tool-call history.
• Preserve native Codex collaboration routing so a Kimi subagent can invoke spawn_agent; descendant requests use the same stateless translation path.
• Keep recursion depth, concurrency, lifecycle, UI, permissions, and sandbox policy under Codex Desktop rather than adding bridge-specific limits.
• Remove the role-unsupported model_supports_reasoning_summaries field from the bundled kimi_frontend template.
• Disable unsupported hosted tools in the safe frontend-role template while retaining multi_agent and multi_agent_v2.
• Extend Rust, Node, and shared compatibility coverage for namespaced functions, namespaced custom tools, history replay, split streaming names, long names, and collisions.
• Version the separately named Node fallback as codex-kimi-bridge-node 0.2.0 for the matching protocol update.

Compatibility boundary

Top-level function and custom tools remain supported. A namespace may contain function and custom children. Other hosted Responses tool types are rejected explicitly and are never dropped silently.

The bridge does not impose extra recursion or concurrency limits. Actual scheduling and any global thread limits remain controlled by Codex Desktop.

No live Kimi request is required for the offline compatibility tests.

Security and installation

The macOS binaries are ad-hoc signed but not Apple-notarized. Verify SHA-256 before bypassing a Gatekeeper warning. The bridge listens on 127.0.0.1 by default and does not log request bodies, credentials, or reasoning content.