memU
personal memory kept as wiki pages, written by an external agent on memU's own jobs. https://github.com/NevaMind-AI/memU
- Core
- 59% · 32.2 / 55 (mean)
- Core (public2)
- public2 run in progress
- Private set
- after the public2 runs
- Full
- 65% · 42.1 / 65 · 65 of 100 measured
- Runs
- 5 · core 32.2–32.2 · the details below are the median run
- Common with gbrain
- 42.2 vs 50.4 on 65
- Chars per answer
- 3,543 · answer after 742
- Write → visible
- 532 ms · p95 532 ms
- Write-back
- 0 of 2 degraded answers came back
- Reached through
- CLI / Python library
- Model on write
- yes: the executor agent (Gemini)
- Key needed
- LLM provider
- Where it runs
- local
- License
- Apache-2.0
- Deterministic
- yes in five builds
Configuration
| Version | memu-cli 0.11.0b3 from source (git main) |
| Mode | self-hosted library, SQLite store, no memu.so account; memU runs no model, an external executor agent writes; run through a JSON-lines bridge |
| Door | progressive_retrieve(question), printed as `memu retrieve` prints it (JSON, indent 2): segments, their files with full content, resources |
| Results | top-5 segments (memU's own retrieve config, not changed), no relevance floor, no cut |
| Model inside | gemini-3-flash-preview (Google, OpenAI-compatible endpoint, function calling) as memU's external executor agent: writes and patches the wiki pages; no model on read |
| Embeddings | gemini-embedding-001, 3072 dimensions (Google, through memU's openai provider with Google's base URL) |
| Storage | SQLite plus a markdown workspace of recall files |
| Not default | executor agent is Gemini with memU's own prompt and five workspace-only file tools instead of a shell (memU leaves the executor to the caller); embeddings through MEMU_EMBED_PROVIDER=openai with MEMU_BASE_URL set to Google; memU's own MEMU_EMBED_MODEL cleared by the bridge; workspace under the store instead of the CLI's fixed one in the home; telemetry, template and docs refresh off (reason not stated); built from source because the PyPI wheels do not ship memorize |
Notes: 0.11.0b3 from source; memU runs no model itself, so Gemini 3 Flash plays the executor agent with memU's prompt and workspace-only file tools (the score is memU plus this executor); gemini-embedding-001. Run 2026-09-30, door retrieve,
fingerprint 1e4728926b. Report:
examples/memu_report/reference.md · All runs: examples/memu_report/repeats.json
Sections
door | 18.8 of 25 · 6 pass · 2 fail | FAIL Which version of MailBridge is installed for the mailbox? — STALE value delivered next to the current one: 1.3.0 FAIL What is the owner's phone number? — missing +39 |
cards | not measured | section not measured: no entity cards in this memory |
updates | 6.7 of 10 · 2 pass · 1 fail | FAIL restating the current value does not pile up a second copy — 2 copies delivered |
time | 0.0 of 10 · 0 pass · 2 fail | FAIL semantic memories carry their age (0/40) — 0% FAIL an imported memory reaches the door with its original date — imported memory not served within 30 s |
live state | 10.0 of 10 · 7 pass · 0 fail | |
abstention | 6.7 of 10 · 0 pass · 4 fail | FAIL What is the Zarpetta module? — 5 keyword facts for something that does not exist FAIL Who is Ottavio Brambillesco? — 5 keyword facts for something that does not exist FAIL Which port does the Fulmicotone service use? — 5 keyword facts for something that does not exist FAIL What does the calendar plugin Girandola do? — 5 keyword facts for something that does not exist |
file search | not measured | section not measured: no repo configured (--repo) |
graph | not measured | section not measured: no case with evidence (all SKIP or no cases) |