Skip to content

fix: MCP recall daemon performance#22

Open
VikingOwl91 wants to merge 1 commit into
qualixar:mainfrom
VikingOwl91:fix/mcp-daemon-recall-performance
Open

fix: MCP recall daemon performance#22
VikingOwl91 wants to merge 1 commit into
qualixar:mainfrom
VikingOwl91:fix/mcp-daemon-recall-performance

Conversation

@VikingOwl91
Copy link
Copy Markdown

Fixes #11.

@VikingOwl91 VikingOwl91 marked this pull request as ready for review May 15, 2026 21:19
@VikingOwl91 VikingOwl91 changed the title [codex] fix MCP recall daemon performance fix: MCP recall daemon performance May 15, 2026
varun369 added a commit that referenced this pull request May 18, 2026
…#22)

- Thread 'fast' recall parameter through the full MCP→daemon→worker
  stack (recall_worker, worker_pool, daemon_proxy, pool_adapter,
  tools_core, tools_v3). Engine.recall(fast=True) shipped in v3.4.40
  but was unreachable from MCP tools until now.

- Wire 'session_id' through pool_recall kwargs (was silently dropped).

- Replace session_init AutoRecall double-call with single pool_recall
  (fast=True) — removes one redundant recall round-trip per session.

- Switch tools_core.recall and tools_v3.recall_trace from hardcoded
  WorkerPool.shared() to choose_pool() — enables daemon-first routing
  (single shared ONNX process instead of N local subprocesses).

- Make FTS triggers idempotent with IF NOT EXISTS (race safety on
  repeated schema init).

- Clean up dead code: remove unused engine= calls in observe and
  close_session; remove unused imports.

Author: VikingOwl91
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

slm remember xxx wait for a long time but seems no response

1 participant