ASM v5.1 is a super memory platform for coding agents: conversation memory + project memory + retrieval/control plane, delivered through a single package and a CLI-first install flow.

@mrc2204/agent-smart-memo provides a project-aware memory layer for OpenClaw plugin/runtime usage, while keeping one shared memory/config model.

Today ASM provides:

• conversation/runtime continuity
• structured slot memory
• semantic retrieval
• graph memory
• project registry + onboarding
• repo-aware indexing / reindexing
• lineage-aware engineering context retrieval
• CLI-based platform install flow for OpenClaw (plus optional OpenCode MCP bridge bootstrap)

This means ASM is best understood as:

a shared memory platform for coding agents, with OpenClaw as the supported plugin/runtime target

ASM has 3 practical layers.

Used for runtime continuity:

• memory_search
• memory_store
• memory_slot_*
• memory_graph_*
• auto-capture / auto-recall

Used for engineering context:

• project registry
• repo root / repo remote identity
• project aliasing
• Jira linkage
• onboarding + index triggers
• lifecycle-aware retrieval boundaries

Used to assemble better context for coding agents:

• semantic recall
• lexical/project filtering
• file/symbol/task lineage
• deterministic project-aware retrieval
• platform install / bootstrap flows

If you only remember one sentence, remember this:

ASM is a super memory platform for coding agents: conversation memory + project memory + runtime delivery in one package.

Primary target today.

Includes:

• OpenClaw plugin entry
• tools / hooks / runtime wiring
• CLI bootstrap flow
• shared ASM config integration

Install:

asm install openclaw

Uses the same package and shared config, with MCP/local runtime wiring.

Install:

asm install opencode

There are currently two supported install flows.

Install the CLI globally first:

npm install -g @mrc2204/agent-smart-memo

Then initialize shared config:

asm init-setup --yes

This creates or updates:

~/.config/asm/config.json

Then install runtime/integration target(s):

asm install openclaw
asm install opencode

If you only want the OpenClaw plugin directly, install it through OpenClaw:

openclaw plugins install @mrc2204/agent-smart-memo

Then continue with OpenClaw-side config/bootstrap as needed.

The command below is not the recommended primary flow right now:

npx @mrc2204/agent-smart-memo install

Use the two supported flows above until CLI bootstrap is fully separated/standardized.

ASM now uses a shared config model.

~/.config/asm/config.json

Runtime/core fields such as:

• projectWorkspaceRoot
• storage.slotDbDir
• llmBaseUrl
• llmApiKey
• llmModel
• embedBaseUrl
• embedBackend
• embedModel
• embedDimensions
• autoCaptureEnabled
• autoCaptureMinConfidence
• contextWindowMaxTokens
• summarizeEveryActions

Compatibility fields (migration/export/rollback tooling window):

• qdrantHost
• qdrantPort
• qdrantCollection
• qdrantVectorSize

Primary wiki-first runtime memory paths should not be treated as Qdrant-dependent.

Bootstrap-safe behavior (Phase 2 bootstrap lane):

• Fresh environments can run capture flow even when distill/LLM extraction capability is unavailable.
• Auto-capture keeps SlotDB runtime-state updates active and applies raw-first fallback capture into wiki raw/live/briefings.
• Deterministic briefing generation remains enabled so startup context stays stable across repeated bootstrap runs.

Isolated continuation distill behavior (Phase 2 continuation-safe lane):

• Auto-capture distill extraction now runs in the isolated continuation lane natively (extractWithIsolatedContinuation) without spawning an external worker process.
• Runtime boundary is explicit 3-layer decomposition:
  • Extractor layer (runExtractorLayerWithLLM + pattern fallback): candidate signals only.
  • Distill layer (runDistillLayer): refine/select/drop/promote hints from candidates.
  • Apply / Memory arrangement layer (DistillApplyUseCase.execute): deterministic writes into SlotDB + wiki lanes (raw, drafts, live, briefings) with loop guards.
• Structured distill contract is supported with optional fields: draft_updates, briefing_updates, log_entries, promotion_hints in addition to slot_updates, slot_removals, memories.
• Native continuation fallback translation is continuation-side: the continuation returns a structured fallback contract (deterministic fallback extraction + diagnostic log_entries) rather than throwing in the hook path.
• Host auto-capture path is boundary-only: build context window -> invoke isolated continuation -> pass returned contract into deterministic apply (no host-side distill mode inference/noise routing).
• Forensic safeguard: same-session direct sendEventSystemMessage/heartbeatnow distill primitive is not used as primary path because auto-capture is attached to agent_end and loop risk is high.

Platform config should stay minimal.

For OpenClaw, ~/.openclaw/openclaw.json should mainly keep:

• enabled
• required runtime fields:
  • projectWorkspaceRoot
  • slotDbDir
  • wikiDir

Example OpenClaw plugin entry:

{
  "enabled": true,
  "config": {
    "projectWorkspaceRoot": "/Users/your-user/Work/projects",
    "slotDbDir": "/Users/your-user/.openclaw/agent-memo",
    "wikiDir": "/Users/your-user/Work/projects/agent-smart-memo/memory/wiki"
  }
}

qmdRoot stays internal and is derived from wikiDir at runtime.

npm install -g @mrc2204/agent-smart-memo
asm init-setup --yes
asm install openclaw --yes

openclaw plugins install @mrc2204/agent-smart-memo

npm install
npm run build
node bin/asm.mjs init-setup --yes
node bin/asm.mjs install openclaw --yes

npm run test:asm-cli
npx tsx tests/test-init-openclaw.ts
npm run build:plugin

ASM supports operator-friendly project onboarding.

/asm_project_index <repo_url>

• resolves repo path/identity when possible
• supports local path import without forced clone
• can reuse an already-registered remote/project identity
• can attach Jira mapping
• can trigger background index flow

Typical path:

1. operator runs /asm_project_index <repo_url>
2. preview shows resolved repo + onboarding choices
3. operator confirms
4. ASM bridges into register / tracker-link / index flow

Relevant areas in the repo include:

• project registry
• onboarding command flows
• background indexing hooks
• lineage-aware retrieval tests

• memory_search
• memory_store
• memory_slot_get
• memory_slot_set
• memory_slot_delete
• memory_slot_list
• memory_graph_*

• project register / list / inspect flows
• project tracker linking
• project indexing / reindexing
• lifecycle-aware retrieval gating
• hybrid lineage context retrieval

• shared config bootstrap
• OpenClaw install flow
• OpenCode install flow
• build/package/publish targets

npm run build

npm run build:plugin
npm run build:core
npm run build:all

npm run package:plugin
npm run package:core

npm run pack:plugin
npm run pack:core

npm run test:asm-cli
npx tsx tests/test-init-openclaw.ts

npm run test:plugin
npm run build:plugin

npx tsx tests/test-project-registry.ts
npx tsx tests/test-project-hybrid-lineage.ts

src/
  adapters/
    openclaw/
  core/
    contracts/
    usecases/
    ingest/
  db/
  hooks/
  services/
  shared/
  tools/

bin/
scripts/
docs/
artifacts/
tests/

A good public-facing description for this repo is:

Agent Smart Memo is a project-aware super memory platform for coding agents, shipped as one package with CLI-first installation for OpenClaw and optional OpenCode MCP integration.

It helps agents:

• remember conversation/runtime state
• store and retrieve structured + semantic knowledge
• onboard and map projects
• index and reindex repos
• assemble better engineering context
• reuse one shared config and one shared memory core across runtimes

MIT © mrc2204