Practical overview
When it fits
Use it when an agent needs persistent memory with controls around untrusted writes, purpose-limited retrieval, and reviewable provenance or audit records.
Who it is for
- Developers integrating persistent memory into AI-agent systems.
- Teams that need to inspect memory provenance and govern which records an agent can retrieve for a given purpose.
Documented uses
- Add governed memory writes and retrieval to an agent system.
- Track the provenance and trust classification associated with stored memories.
- Review hash-chained audit events for memory operations.
- Run a local demo stack or explore sample governance flows in a hosted sandbox.
Requirements & limits
The repository identifies the project as open source under the MIT license. Its hosted demo is described as requiring no account or local setup and is intended for synthetic data; local quickstart requires Docker.
- The documented local stack uses Postgres with pgvector alongside a REST API and web console.
- The write path includes injection scanning and deduplication; retrieval combines hybrid search with a privilege gate and purpose-binding policy checks.
- The project describes provenance lineage, quarantine and deletion operations, and a hash-chain audit verifier.
- The hosted sandbox is for synthetic demo data, not personal, customer, production, or secret information.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Explore governed writes and retrieval in the hosted sandbox
Inspect how the demo classifies sample writes, filters retrieval, and records governance events.
Before you start
- A web browser and access to the hosted demo.
- Use synthetic demo data only; the sandbox may be reset.
Steps
Open https://demo.metaworkers.ai/ and choose Write, then select Benign example.
Choose Phishing via a "trusted" source or Fake system override and review the resulting untrusted or quarantined label.
Use Search with its default option to see that untrusted or quarantined records are excluded; enable Include untrusted/quarantined only to inspect those records explicitly.
Open Audit Log to inspect the related governance event.
Expected outcome
A walkthrough of sample memory writing, governed retrieval, and the corresponding audit event.
Guide limitations
- The hosted deployment is a synthetic-data demo; do not enter personal, customer, production, or secret data.
- The demo has no anonymous reset button, and its records may be reset by the deployment owner.
Start the seeded local demo with Docker
Bring up the local Postgres-plus-pgvector store, REST API, web console, and seeded demo tenant.
Before you start
- A clean clone of the repository.
- Docker installed; no Python environment or conda setup is required.
- Windows PowerShell, macOS, or Linux.
Steps
Clone https://github.com/Metaworkers-ai/governedmemory.git and change into the governedmemory directory.
On macOS or Linux, run ./scripts/quickstart.sh; in Windows PowerShell, run ./scripts/quickstart.ps1.
Open the Web console URL printed by the wrapper, then use Write to submit a sample and Audit Log to inspect its event.
Expected outcome
A local seeded demo stack with a web console for trying a governed write and inspecting its audit event.
Guide limitations
- Docker must be installed; the wrapper exits without changing anything if Docker is unavailable.
- The first run builds the API and web images and may take about five minutes on a normal connection.
- The sample outcome is part of the documented demo flow, not a compatibility or security guarantee.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- Metaworkers-ai/governedmemory · Official homepage title · retained in its source language