fix(memories): get memory tool, mem0 integration update

[icecrasher321]

Summary

• Memory block -- did fuzzy match on conversation id instead of directly getting it. Causing some edge case bugs.
• Mem0 -- endpoints are out of date

Type of Change

• Bug fix

Testing

Tested manually

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		May 3, 2026 1:40am

[cursor]

PR Summary

Medium Risk
Updates both internal memory endpoints and the Mem0 integration request/response shapes; mistakes could cause missing memories or broken third-party calls. Changes are localized but touch user-facing retrieval and external API behavior.

Overview
Fixes internal memory lookups/deletes to exclude soft-deleted rows by adding isNull(memory.deletedAt) filters in /api/memory/[id] and /api/memory delete-by-query.

Updates the Mem0 integration to current v3 endpoints and payloads, adds shared validation via parseMem0Messages, introduces page support for get/list, and adjusts tool/block outputs to match new response envelopes (including queued add-memory fields and paginated get metadata). Adds/updates focused Vitest coverage for Mem0 tools/block and for memory_get using exact /api/memory/:id paths; trims a flaky Sharp test by removing the small-image case.

^{Reviewed by Cursor Bugbot for commit 21ef4ae. Configure here.}

[greptile-apps]

Greptile Summary

This PR fixes two bugs: the memory get tool now performs an exact path-based lookup (/api/memory/{id}) instead of a fuzzy query-parameter search, and the Mem0 integration updates all three tools (add, get, search) from deprecated v1/v2 endpoints to v3 equivalents with matching request/response shapes.

• P1 — DELETE and PUT handlers in route.ts do not include isNull(memory.deletedAt) in their WHERE clauses, so soft-deleted records that are invisible to GET can still be mutated or hard-deleted, breaking soft-delete semantics.
• P1 — transformResponse in memory/get.ts calls response.json() without checking response.ok, silently converting any HTTP error (including 404) into { success: true, memories: [] } and hiding real failures from callers.

Confidence Score: 3/5

Not safe to merge as-is — two P1 bugs exist: soft-delete inconsistency in route.ts and silent HTTP error masking in the memory get tool.

Two P1 findings cap the score at 4; the soft-delete inconsistency affects correctness of write operations on the same route that was just fixed for reads, pulling the score below the ceiling to 3.

apps/sim/app/api/memory/[id]/route.ts (DELETE and PUT soft-delete filters), apps/sim/tools/memory/get.ts (HTTP error handling in transformResponse)

Important Files Changed

Filename	Overview
apps/sim/app/api/memory/[id]/route.ts	Adds isNull(memory.deletedAt) to GET to fix soft-delete visibility, but DELETE and PUT existence checks still omit this filter, creating inconsistent behavior for soft-deleted records.
apps/sim/tools/memory/get.ts	Correctly switches from fuzzy query-based lookup to exact path-based lookup via encodeURIComponent; transformResponse silently converts API HTTP errors into empty success results.
apps/sim/tools/mem0/add_memories.ts	Updates to v3 add endpoint; refactors message parsing and response handling to match new queued-processing API shape; includes trimming userId.
apps/sim/tools/mem0/get_memories.ts	Upgrades to v3 list endpoint with proper filter structure; adds URL encoding for memoryId; pagination hardcoded to page 1 with no way to retrieve subsequent pages.
apps/sim/tools/mem0/search_memories.ts	Updates to v3 search endpoint; replaces ad-hoc response mapping with type-safe helpers; correctly wraps results from the new envelope format.
apps/sim/blocks/blocks/mem0.ts	Adds parseMem0Messages helper with strict role validation; removes deprecated default userId value; moves date range fields to advanced mode; simplifies params/switch logic.
apps/sim/tools/mem0/types.ts	Adds typed param and response interfaces for all three operations; updates ADD_MEMORY_OUTPUT_PROPERTIES to reflect new async/queued response shape.

Comments Outside Diff (2)

1. apps/sim/app/api/memory/[id]/route.ts, line 131-135 (link)

   Soft-deleted records not filtered in DELETE and PUT handlers

   The GET handler now correctly excludes soft-deleted rows via isNull(memory.deletedAt), but the existence checks in the DELETE (line 131–135) and PUT (line 183–191) handlers still query without that filter. A soft-deleted memory will be invisible to GET (returning 404) yet still matched by DELETE and PUT — breaking the invariant that deletedAt != null means "does not exist". The PUT handler will update and return a soft-deleted record; the DELETE handler will hard-delete it, bypassing soft-delete semantics entirely. Both handlers should add isNull(memory.deletedAt) to their existence-check (and update/delete) WHERE clauses.

2. apps/sim/tools/memory/get.ts, line 49-61 (link)

   HTTP error responses silently become empty success results

   When the API returns a 404 (memory not found), the response body is { success: false, error: { message: 'Memory not found' } }, so result.data is undefined. The !memory branch then returns { success: true, output: { memories: [], message: 'No memories found' } }, hiding the fact that the request actually errored. Callers have no way to distinguish "memory doesn't exist" from "API returned an error". Checking response.ok first makes the failure explicit.

_{Reviews (1): Last reviewed commit: "fix(memories): get memory tool, mem0 int..." | Re-trigger Greptile}

[icecrasher321]

bugbot run

[icecrasher321]

bugbot run

[icecrasher321]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 21ef4ae. Configure here.}