JDSS/mem0
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(deepseek-plugin): first-turn recall, native activation, and cancel-safe recall (#7284, #7302)

Browse Source 
Ports #7285 onto current main and adds a canDeliver() gate so a canceled recall does not mark undelivered memories as seen.

Refs #7284, #7302. Supersedes #7285.
This commit is contained in:
Himanshu-Sangshetti
2026-10-05 09:48:32 +05:30
parent abb81c88e1
commit df7cd065da
11 changed files with 313 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files
integrations/deepseek-plugin/README.md
+24 -7
View File
@@ -13,7 +13,7 @@ It gives a Harness agent automatic long-term memory plus two explicit memory too
Unlike the local/file-based memory plugins in the ecosystem, Mem0 is a managed backend: server-side extraction, semantic dedup and conflict resolution, and memories that other agents can retrieve when their user and entity filters match.
Current package version: `0.3.0`.
Current package version: `0.3.2`.
Sidekick is available only in the [Claude Code plugin](../claude-code-plugin/README.md#sonnet-sidekick-agent).
@@ -47,21 +47,27 @@ Cordis owns listener and tool cleanup when the plugin unmounts. Every automatic
mkdir -p /tmp/mem0-deepseek-plugin
pnpm pack --pack-destination /tmp/mem0-deepseek-plugin
```
2. Set your Mem0 key:
2. Set your Mem0 key, a stable user identity, and a funded model API key:
```sh
export MEM0_API_KEY=...
export MEM0_USER_ID=your-user-id
export DEEPSEEK_API_KEY=...
```
3. Install it into a disposable Harness profile:
3. Install it into a disposable Harness profile (requires pnpm 10 or newer; pnpm 9 ignores the profile's `pnpm-workspace.yaml` settings and the install fails on an unpublished internal package):
```sh
DSH_HOME=/tmp/mem0-dsh-dev pnpm dlx @deepseek-ai/dsh@0.1.1-rc.2 \
plugin --profile headless add /tmp/mem0-deepseek-plugin/mem0-deepseek-plugin-0.3.0.tgz
plugin --profile web add /tmp/mem0-deepseek-plugin/mem0-deepseek-plugin-0.3.2.tgz
```
4. Copy `cordis.example.yml`, set its installed package path and your `userId`, then run Harness with the same profile:
4. Run Harness with the same profile. The installed bundle activates Mem0 automatically:
```sh
DSH_HOME=/tmp/mem0-dsh-dev pnpm dlx @deepseek-ai/dsh@0.1.1-rc.2 \
web --patch ./integrations/deepseek-plugin/cordis.example.yml
web
```
5. Open http://127.0.0.1:3080 and ask the agent to remember something, then recall it in a later turn.
5. Open http://127.0.0.1:3080, select a workspace, and state a synthetic preference without mentioning Mem0. Wait for asynchronous extraction to finish, then start a fresh session and ask for that preference without tools. Inspect the `mem0:recall` context to verify automatic recall.
Build from the full repository: the source imports the sibling shared core. `pnpm pack` runs the build and includes the activation patch. An installed tarball is self-contained. For custom settings, copy `cordis.example.yml` and pass its absolute path with `--patch`. When upgrading from the old manual setup, remove the old patch that **inserts** a `mem0` row; the bundle now inserts it.
On macOS, set `CHOKIDAR_USEPOLLING=1` if Harness reports `EMFILE`. The Mem0 key and model-provider key are separate credentials.
For a Mem0 Platform on-prem or dedicated deployment, point `config.host` at that base URL (defaults to `api.mem0.ai`). `host` overrides the Platform base URL. It does not support the self-hosted Mem0 OSS API.
@@ -71,6 +77,7 @@ For a Mem0 Platform on-prem or dedicated deployment, point `config.host` at that
|---|---|---|---|
| `apiKey` | no | `$MEM0_API_KEY` | Mem0 platform API key |
| `userId` | yes | | Entity that owns the memories |
| `memoryScope` | no | `user` | `user` shares memory across workspaces; `workspace` isolates all automatic and explicit operations by the session workspace |
| `allowUserOverride` | no | `false` | Permit model-selected access to a different user only in a trusted multi-user deployment |
| `host` | no | `api.mem0.ai` | Platform base URL (on-prem / dedicated) |
| `autoRecall` | no | `true` | Recall relevant memory before model requests |
@@ -80,10 +87,20 @@ For a Mem0 Platform on-prem or dedicated deployment, point `config.host` at that
Automatic capture and recall use the configured `userId` across sessions. Automatic writes do not attach a repository ID or `runId`.
This cross-workspace sharing is intentional compatibility behavior. Set `memoryScope: workspace` in the profile patch to isolate memory. The plugin then adds an `appId` derived from the canonical absolute session workspace path to every write and the matching `app_id` to every search. Symlink aliases share a scope; different directories (including separate clones or worktrees) do not. Moving a workspace changes its scope. Tools cannot override it. Missing or invalid workspace paths skip automatic memory operations and reject explicit tools, without falling back to user-wide access.
Workspace scope does not migrate existing user-only memories. User scope still searches all memories for that user, including workspace-tagged memories; isolation applies when the plugin is configured with workspace scope. These filters are application-level separation, not separate Mem0 credentials.
Both `search_memory` and `add_memory` accept optional `agentId` and `runId`. On search, these narrow the returned memories; on add, they attach those identities to the stored memory. Pass a known `runId` to search memories explicitly saved with that session ID. This does not include automatically captured user-only memories or identify the session making the request.
Per-call `userId` overrides are rejected unless the operator enables `allowUserOverride: true`. Automatic recall and capture always use the configured user.
## Extraction and recall limits
The plugin sends full redacted completed turns to Mem0; Mem0 extracts the memories asynchronously. A queued write is not proof that every extracted fact is already searchable. Even an initial nonempty memory list can be incomplete. Check the stored memories after processing before diagnosing extraction loss.
Automatic recall is a bounded first pass (up to five results, 4,000 context characters, and a two-second wait). `search_memory` uses a focused query and returns up to ten results by default. Exact preference extraction remains model-dependent; include filenames and numeric constraints in release tests, and distinguish standing preferences from one-off requests.
## Telemetry
Writes are tagged `source="DEEPSEEK_HARNESS"`. That value has to exist in the backend's `EventSource` enum for usage to surface by name; until it does, these writes read as `OTHERS`. It is added by [mem0ai/platform#3602](https://github.com/mem0ai/platform/pull/3602), which has to ship before this claim is true.