agentmemory (Gemini)
the same memory with its LLM features on: compression, graph extraction, consolidation. https://github.com/rohitg00/agentmemory
- Core
- 65% · 35.5 / 55 (mean)
- Core (public2)
- public2 run in progress
- Private set
- after the public2 runs
- Full
- 66% · 42.6 / 65 · 65 of 100 measured
- Runs
- 5 · core 35.5–35.5 · the details below are the median run
- Common with gbrain
- 42.6 vs 50.4 on 65
- Chars per answer
- 6,338 · answer after 362
- Write → visible
- 253 ms · p95 300 ms
- Write-back
- 0 of 0 degraded answers came back
- Reached through
- MCP / REST
- Model on write
- yes: compression and graph extraction (Gemini)
- Key needed
- LLM provider
- Where it runs
- local
- License
- Apache-2.0
- Deterministic
- yes in five builds
Configuration
| Version | @agentmemory/agentmemory 0.9.29 (npm) with iii-engine 0.11.2 |
| Mode | with Gemini: LLM features on (compression, graph extraction, consolidation) |
| Door | memory_recall(query, limit=10) through POST /agentmemory/mcp/call, the route the MCP stdio shim forwards to; the tool's own text |
| Results | limit 10 (its default), format full (its default), no relevance floor, no cut |
| Model inside | gemini-3-flash-preview (Google): compresses observations, extracts the knowledge graph, runs consolidation |
| Embeddings | gemini-embedding-001, 768 dimensions (Google, the one Gemini embedder it ships) |
| Storage | local scratch data directory on the iii engine (format not stated) |
| Not default | GEMINI_MODEL=gemini-3-flash-preview and EMBEDDING_PROVIDER=gemini (the Gemini setup); AGENTMEMORY_AUTO_COMPRESS=true, GRAPH_EXTRACTION_ENABLED=true, CONSOLIDATION_ENABLED=true (the three flags /config/flags lists as needing an LLM); the import also runs /graph/build and /consolidate-pipeline |
Notes: npm 0.9.29, gemini-3-flash-preview, gemini-embedding-001; same door and set as the keyless row. Run 2026-09-30, door recall,
fingerprint 34483ae0a3. Report:
examples/agentmemory_report/reference_gemini.md · All runs: examples/agentmemory_report/repeats_gemini.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 — 3 copies delivered |
time | 6.7 of 10 · 4 pass · 2 fail | FAIL semantic memories carry their age (38/49) — 78% FAIL an imported memory reaches the door with its original date — served with no date |
live state | 7.1 of 10 · 5 pass · 2 fail | FAIL after overwriting the canary the door serves the new value, never the old one — overwrite-to-visible 315 ms · STALE: old and new value delivered together FAIL two writes in quick succession: the door settles on the second, never back on the first — second write visible in 270 ms · 24 read(s) served the first write after the second was already visible |
abstention | 3.3 of 10 · 0 pass · 4 fail | FAIL What is the Zarpetta module? — episodes presented as direct matches; 6 keyword facts for something that does not exist FAIL Who is Ottavio Brambillesco? — episodes presented as direct matches; 6 keyword facts for something that does not exist FAIL Which port does the Fulmicotone service use? — episodes presented as direct matches; 6 keyword facts for something that does not exist FAIL What does the calendar plugin Girandola do? — episodes presented as direct matches; 6 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) |