OpenTrust

OpenTrust is the proposed official memory layer standard for OpenClaw.

It is a local-first, operator-grade memory and traceability system designed to make OpenClaw data:

• reliably captured
• durably stored
• safely retrieved
• richly linked
• explainable
• measurable
• operationally useful

OpenTrust is not just a search feature. It is the memory substrate that should let OpenClaw answer questions like:

• what happened?
• what evidence supports that answer?
• what tool, workflow, or event caused the outcome?
• what should be remembered long-term?
• what changed over time?
• what is stale, missing, uncertain, or risky?
• what insights can be derived from the full event stream?

Product position

OpenTrust should be treated as:

• the OpenClaw memory layer reference implementation
• the architecture blueprint for storage + retrieval + traceability
• the standard model for durable evidence, insights, lineage, and metrics
• the foundation for future OpenClaw memory APIs, operator UX, and agent-facing retrieval

It should not be framed as a detached product thesis competing with OpenClaw. Its primary purpose is to strengthen OpenClaw itself.

Core standard goals

The OpenClaw memory layer should provide:

1. Reliable storage

• append-safe ingestion of events and evidence
• durable local system of record
• schema evolution and migration support
• reproducible IDs and stable references
• auditable persistence

2. Reliable retrieval

• lexical retrieval for exact matches
• semantic retrieval for fuzzy recall
• structured SQL and relational joins for investigations
• lineage-aware lookup across sessions, workflows, artifacts, and capabilities
• ranking that prefers evidence over unsupported summaries

3. Explainable traceability

• event lineage
• tool-call / tool-result pairing
• workflow step attribution≠
• artifact provenance
• ingestion freshness and cursor state
• clear distinction between observed evidence and inferred conclusions

4. Dynamic intelligence

• memory-aware search
• derived summaries
• anomaly surfacing
• change detection
• operator health views
• metrics and insight generation over time

5. OpenClaw integration

• interoperable with OpenClaw session traces
• interoperable with cron/workflow runs
• compatible with future agent memory APIs
• suitable for direct use by OpenClaw UI surfaces and agent tools
• extensible toward official memory contracts and retrieval services
• planned for first-class packaging as an OpenClaw plugin

Current implementation status

OpenTrust already includes:

• polished dashboard-first UI shell
• local SQLite runtime
• explicit bootstrap / ingest / query architecture
• OpenClaw session ingestion
• cron/workflow ingestion
• ingestion state tracking
• FTS-backed investigation search
• artifact extraction and registry population
• trace detail pages
• workflow detail pages
• artifact explorer
• saved investigations
• semantic chunk indexing
• sqlite-vec activation
• secret-blocking pre-commit protection

Strategic direction

The correct long-term direction is:

• keep OpenTrust’s memory model, evidence framing, lineage, and operator UX
• align all roadmap language around OpenClaw memory infrastructure
• treat any future backend shifts as implementation choices, not product redefinitions

That means:

• SQLite-first local durability remains valuable
• other runtimes or providers may be added later
• but the governing idea is the OpenClaw memory layer standard

Current execution priority

OpenTrust is currently in a pre-prerequisite holding pattern.

The upstream priority order for memory-layer work is:

1. persistence
2. reliable run completion

Desktop application packaging (Tauri) can proceed in parallel — see docs/DESKTOP-APPLICATION-PLAN.md.

Until persistence and run completion are dependable, memory-layer expansion should focus on:

• docs alignment
• scope guardrails
• low-risk cleanup
• preserving a truthful baseline

Memory-layer expansion should not pretend to outrank runtime reliability work.

Principles

• OpenClaw-first
• local-first durability
• evidence-backed answers
• progressive disclosure
• traceability over vibes
• operator-grade reliability
• memory as infrastructure, not garnish

Getting started

The fastest way to set up OpenTrust is the interactive setup script. It walks through every step, explains what each one does, detects prior work, and lets you skip anything:

./scripts/setup.sh

Prerequisites

• Node.js 18+
• pnpm — install with npm install -g pnpm or see pnpm.io
• git — required for Husky pre-commit hooks

Manual setup

If you prefer to run each step yourself:

pnpm install                  # install deps + set up Husky hooks
cp .env.example .env          # create local env (optional overrides)
pnpm run db:init              # create/migrate SQLite database
pnpm run ingest:openclaw      # import OpenClaw session transcripts
pnpm run ingest:cron          # import cron/workflow run history
pnpm run index:semantic       # build semantic chunk + vector index
pnpm dev                      # start the dev server on localhost:3000

Environment variables

Optional overrides are documented in .env.example.

Variable	Purpose
OPENTRUST_AUTH_MODE	Auth mode for the app surface: token, password, or none.
OPENTRUST_AUTH_TOKEN	Shared credential used when OPENTRUST_AUTH_MODE=token.
OPENTRUST_AUTH_PASSWORD	Shared credential used when OPENTRUST_AUTH_MODE=password.
OPENTRUST_ALLOW_LOCALHOST_BYPASS	Allow bypass on localhost / 127.0.0.1 / ::1. Set false for strict auth everywhere.
OPENTRUST_APP_URL	Canonical app URL used for metadata / Open Graph resolution.
OPENTRUST_SQLITE_VEC_PATH	Override path to the sqlite-vec extension. Leave unset unless auto-detection fails.

Security modes

Recommended defaults:

• Local-only development: OPENTRUST_AUTH_MODE=token with OPENTRUST_ALLOW_LOCALHOST_BYPASS=true
• Strict protected mode: OPENTRUST_AUTH_MODE=password with OPENTRUST_ALLOW_LOCALHOST_BYPASS=false
• Unprotected mode: OPENTRUST_AUTH_MODE=none only for isolated localhost experimentation

OpenTrust authenticates at the app/server boundary before protected evidence is rendered. The browser does not authenticate directly to SQLite.

Auth endpoints also enforce same-origin checks for POST requests, and the auth cookie is issued as an httpOnly, SameSite=Strict, time-bounded session.

Available scripts

Command	Description
pnpm dev	Start the Next.js dev server
pnpm run db:init	Create/migrate the SQLite database
pnpm run ingest:openclaw	Import recent OpenClaw sessions
pnpm run ingest:cron	Import cron workflow run history
pnpm run index:semantic	Rebuild semantic chunks + vector index
pnpm run typecheck	Run tsc --noEmit
pnpm run secrets:check	Scan the repo for leaked secrets

Safety

This repo includes a pre-commit secret scan:

• Husky pre-commit hook
• Secretlint ruleset

OpenTrust is intended for localhost-only use. The /api/memory/* routes are unauthenticated by design and should not be exposed to untrusted networks.

Run manually with:

pnpm run secrets:check

Docs

• docs/ARCHITECTURE.md
• docs/FIELD-MANUAL.md
• docs/INGESTION.md
• docs/PHASES.md
• docs/EXECUTION-PIPELINE.md
• docs/SCOPE-GUARDRAILS.md
• docs/CONVEX-PRODUCT-SPEC.md
• docs/CONVEX-REPLACEMENT-BLUEPRINT.md
• docs/MEMORY-LAYER-STANDARD.md
• docs/MEMORY-API-CONTRACT.md
• docs/CURATED-MEMORY-DESIGN.md
• docs/OPENCLAW-MEMORY-SOURCES-AND-BACKUPS.md
• docs/CRM-SCAFFOLDING-NOTES.md
• docs/PHASE-9-IMPLEMENTATION-PLAN.md
• docs/DESKTOP-APPLICATION-PLAN.md
• docs/OPENCLAW-PLUGIN-PACKAGING.md
• docs/PLUGIN-READY-REFACTOR-PLAN.md
• docs/OPENCLAW-PLUGIN-DRAFT-ARTIFACTS.md
• docs/POST-PREREQUISITE-PR-BACKLOG.md
• docs/UI-UX-REVAMP-PLAN.md
• plugin-drafts/opentrust/
• docs/PLUGIN-DRAFT-RUNTIME-MAP.md
• docs/PLUGIN-PROMOTION-CHECKLIST.md
• docs/PLUGIN-CHANGE-TRACKER.md
• docs/DOCS-ALIGNMENT-NOTES.md
• db/0001_init.sql
• SECURITY.md