Skip to content

Reality check & status

This page answers one question: what is actually shipped, and how quickly can you verify it yourself?

Legend

  • DONE — works locally and deterministically (no network required unless stated)
  • PARTIAL — shipped, but optional/experimental and/or depends on external services
  • ROADMAP — planned; not shipped yet

Quick reality check (verifiable in ~60s)

1) Install deps (from a fresh clone)

uv sync --locked
uv run --python 3.13 --frozen -- python -m openclaw_mem --help

Note: this repo defines the openclaw-mem console entrypoint, so uv run openclaw-mem ... is supported after dependency install (uv sync --locked). The docs use uv run --python 3.13 --frozen -- python -m openclaw_mem ... as the most explicit source-checkout form.

2) Local DB smoke test (no OpenClaw required)

DB=/tmp/openclaw-mem-realitycheck.sqlite

uv run --python 3.13 --frozen -- python -m openclaw_mem init --db "$DB" --json

cat > /tmp/openclaw-mem-sample.jsonl <<'JSONL'
{"ts":"2026-02-12T08:10:00Z","kind":"tool","tool_name":"web_search","summary":"searched docs status markers","detail":{"query":"status markers DONE PARTIAL ROADMAP"}}
{"ts":"2026-02-12T08:11:00Z","kind":"note","summary":"Docs updated","detail":{"path":"README.md"}}
JSONL

uv run --python 3.13 --frozen -- python -m openclaw_mem --db "$DB" --json ingest --file /tmp/openclaw-mem-sample.jsonl
uv run --python 3.13 --frozen -- python -m openclaw_mem recall "Docs" --db "$DB" --mode auto --json
uv run --python 3.13 --frozen -- python -m openclaw_mem pack --db "$DB" --query "Docs status" --trace --json
uv run --python 3.13 --frozen -- python -m openclaw_mem curate scan --target memory --db "$DB" --json
uv run --python 3.13 --frozen -- python -m openclaw_mem db info --db "$DB" --json

Expected output shape (minimal):

  • init returns openclaw-mem.init.v1 plus config/capability evidence.
  • ingest returns something like:
  • { "inserted": 2, "ids": [1,2] }
  • recall reports the selected/fallback lane and returns attributable results.
  • pack returns bounded items, citations, composite score evidence, and a trace.
  • curate scan is review-only and emits governed candidates without applying a mutation.
  • db info reports schema, FTS, taxonomy/lifecycle distributions, embeddings, and optional sqlite-vec posture.

3) Engine receipt debug smoke (local-only, no memory text)

node --experimental-transform-types tools/mem-engine-receipts-debug.mjs

Expected: - prints one synthetic recall lifecycle receipt (openclaw-mem-engine.recall.receipt.v1) - prints one synthetic autoCapture lifecycle receipt (openclaw-mem-engine.autoCapture.receipt.v1) - payload contains IDs/scores/counts only (no memory content text)

Status map (operator view)

Core (local, deterministic) — DONE

  • SQLite ledger + FTS5
  • JSON receipts (--json)
  • Progressive recall: ingest → search → timeline → get
  • Context pack command (pack) with fail-open baseline behavior
  • Primary recall router across lexical/vector/hybrid/graph lanes
  • Six-state lifecycle, eight-kind taxonomy, pack quotas, composite scoring, citation-only use tracking, and reversible soft archive
  • Exact sqlite-vec → NumPy → Python vector fallback chain with explicit receipts
  • Deterministic graph-preflight pack integration: pack --use-graph=off|auto|on with traceable trigger/probe receipts
  • Pack decision surfaces (trust_policy, graph provenance_policy, policy_surface, lifecycle_shadow) with bounded trace receipts
  • Deterministic triage for automation (triage --mode heartbeat|cron-errors|tasks)

Quality layers — PARTIAL

  • Embeddings + vector search (embed, vsearch) — API and optional local FastEmbed providers are supported; missing optional providers degrade explicitly
  • Hybrid retrieval (hybrid) — shipped; optional rerank/network providers remain opt-in
  • AI compression (summarize) — requires an API key
  • Graph query plane (graph topology-refresh, graph query ..., graph health, drift/provenance checks) — shipped foundation; deeper integrations still evolving
  • Graph semantic match v0 (graph match "<idea/query>") — shipped local-first idea → project slice with explanation paths; deeper typed graph automation still evolving
  • Graph-assisted search first slice (search --graph --graph-path <graph.json>) — shipped opt-in; default off, fail-open to lexical-only, deterministic blended ranking with rank_components trace
  • Graph-aware pack ranking (pack --graph-aware) — shipped opt-in trace lane; default off, fail-open, zero-write
  • Topology writer (graph render topology) — shipped generated review surface (Markdown summary + Mermaid view + drift report); generated docs are review-only, not source of truth
  • Graph fact guard pilot (graph fact propose|guard) — shipped advisory-only pre-edit guard; proposals require source refs, stale/superseded facts are excluded from current guard truth
  • Recommendation-only optimization observers (optimize review, optimize consolidation-review) — shipped in the current release scope; now include recent-use-aware decay protection plus bounded importance-drift spot checks while still proposing only and never mutating stored memories
  • Dual-language fields (--text-en, EN embedding table) — shipped, still evolving
  • Episodic event capture/ingest/query lane — shipped foundation; operator flows still evolving
  • Episodic verbatim semantic lane (episodes embed, episodes search --mode hybrid|vector) — shipped first production slice; still operator-driven and read-only

OpenClaw integration — PARTIAL

  • Auto-capture plugin (extensions/openclaw-mem) — captures tool results to JSONL
  • Backend-aware annotations (records backend + memory tool actions for observability)
  • Gateway-assisted semantic recall (Route A): index + semantic — depends on OpenClaw gateway + memory_search
  • Host bridge envelopes (bridge status|recall|store) — per-operation versioned JSON envelopes (openclaw-mem-engine.bridge.<op>.v1); recall is read-only, store performs a canonical write only with explicit operator approval
  • Unified installer/doctor adapters for Claude Code, Codex, OpenClaw, generic, Gemini CLI, Cursor, and Windsurf
  • MCP mem_recall/mem_pack equivalence plus read-only graph tools

Near-term roadmap — ROADMAP

  • Graph roadmap depth: richer typed-entity wiring + deeper operator queries/autonomy on top of the shipped query plane
  • Topology seed automation (topology-seed) so curated graph truth is easier to bootstrap and maintain
  • Consolidate/promotion/dissent workflows on top of v2 lifecycle and scoring evidence

Graphic Memory compiled synthesis — PARTIAL

  • graph synth compile — compile a reusable synthesis card from explicit refs or query-preflight selection
  • graph synth stale — deterministic stale check (source digest + query-selection drift)
  • graph lint — deterministic health checks for stale cards / missing source metadata / unreferenced capture rows
  • Graph preflight preference for fresh synthesis cards when they cover multiple selected raw refs
  • Graph pack preference for fresh synthesis cards when explicit refs are covered by a fresh synthesis card
  • Main pack --use-graph now records graph-consumption receipts and elides raw L1 lines already covered by preferred synthesis cards in the combined graph-aware bundle
  • cmd_hybrid now prefers fresh synthesis cards in top results when they cover multiple high-ranked raw hits, and emits graph-consumption receipts on the synthetic result
  • search now prefers fresh synthesis cards in top results when multiple matched raw hits are covered by the same card, with graph-consumption receipts on the synthetic result
  • graph synth refresh now replays the old card’s selection, emits a fresh replacement card, and marks the old card as superseded with a superseded_by receipt
  • graph lint now reports deterministic coverage pressure / candidateCardSuggestions using scope + repeated-keyword clusters, not just scope-only grouping
  • graph synth stale / graph lint now surface deterministic review + contradiction-keyword signals from newly selected refs
  • Optional Markdown materialization during compile (--write-md)