Repository navigation
feat(memory): batch learning and surface cited project facts - #94
Merged
Merged
Conversation
|
ScriptedAlchemy
enabled auto-merge (squash)
November 9, 2026 02:15
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Automatic memory previously scheduled a model call for each settled run, and the composer required an explicit search before it could show a relevant fact.
The server now keeps a durable queue of new runs by project and thread family. It reviews up to eight at a time after eight completions or 30 minutes of inactivity. Short acknowledgements and unsupported assistant-only exchanges can skip generation. Candidates need canonical citations to a user statement or distinct evidence across completed runs in different threads. Repeated facts gain verified sources while preserving their content, origin and trust. Existing facts and old threads require no backfill.
Web, Electron and native mobile show up to two cited matches while composing a first request. Use opens the selection for review and attachment. Dismiss leaves the draft alone. Passive previews search facts without model calls, transcript search, search receipts or trust changes. Clients check the connected server’s capability before sending draft text. Eligible unattended first turns can receive two opaque fact IDs with a 150 ms lookup deadline.
Regression coverage exercises the durable queue, rollback, source attribution, context budgets, older servers, and the web and native suggestion flow. The changed-file gate passed, including 360 tests, the React compiler checks and typechecking 36 affected packages. The advertised MCP tool list shrank by 18 bytes. Two MCP recall eval requests were assembled in dry-run mode. The web captures use the Chromium component harness. The T3 Browser and Device panel tools were unavailable for installed-client verification.
Model and harness: GPT-6 and GPT-6.1 Sol subagents in Codex.