v0.4.13

fix(runtime): remove memory_mb from _COMPAT_FIELDS - eliminates BoxSessionConflictError when python and node MCPs share mcp-shared session with different memory requirements

v0.4.12

fix: add memory_mb to BoxManagedProcessSpec so per-process memory override works; eliminates AttributeError and node/npx MCP OOM kills (return_code=137)

v0.4.11

fix: add memory_mb to BoxManagedProcessSpec + per-process memory override in nsjail backend - eliminates node/npx MCP OOM kills

v0.4.10

fix: per-process memory_mb for node/npx MCP OOM fix

0.4.9

Raise nsjail RLIMIT_AS to hard (#97) so node/npx stdio MCP servers (e.g. firecrawl-mcp) no longer crash with WebAssembly.instantiate OOM. Physical memory stays bounded by cgroup_mem_max / container mem_limit.

0.4.8

Fix long-lived Box stdio MCP servers dropping with Box managed process exited unexpectedly / Failed after 4 attempts.

• Server-driven WebSocket heartbeat (30s) on the managed-process stdio relay (#95), preventing transient WS drops caused by a missed client ping/pong under event-loop stalls.

Pairs with LangBot#2303 which makes the LangBot side reuse the live managed process on a transport reconnect instead of tearing it down and rebuilding.

0.4.7

Release langbot-plugin 0.4.7.

0.4.6

Release langbot-plugin 0.4.6.

0.4.5

What's Changed

• fix(entities): add provider_specific_fields to Message / ToolCall / MessageChunk / ToolCallChunk (#83)

  Lets provider-specific metadata round-trip through the message entity layer. Specifically required for Gemini's thought_signature, which must be preserved across tool-call rounds for function calls to work correctly (see langbot-app/LangBot#1899). Optional field defaulting to None → fully backward-compatible.

Full Changelog: 0.4.4...0.4.5

0.4.4

feat(runtime): capture per-plugin stderr logs into a ring buffer and expose them via a new GET_PLUGIN_LOGS action, so LangBot can display plugin logger output on the plugin detail page.