DevMemo AI is a self-hosted developer knowledge base for individuals and small teams. It is built on Memos, which remains the source of truth for original Memo data, user identities, and visibility permissions. A separate FastAPI AI Service stores AI-derived state only; it does not introduce a second identity or authorization system.
Use DevMemo AI to capture code snippets, bug reports, solutions, and technical decisions as traceable developer knowledge. After human review, accepted insights can be compiled into bounded, copyable Context Packs for an IDE or a subsequent tool.
Unofficial downstream project. DevMemo AI is not affiliated with, endorsed by, or supported by the Memos project. Read UPSTREAM.md and NOTICE for the upstream baseline, ownership boundary, and attribution.
- Keep Memo data, users, and permissions in Memos; AI state never replaces the Memos source of truth.
- Generate structured templates and AI-derived insights for Code Snippets and Bug Reports.
- Review and accept or reject insights in the Memo detail view before they can contribute to a working context.
- Build
context-pack-v1from explicitly visible Memos and accepted insights. It contains only safe titles, summaries, and source references—not raw Memo content, Webhook payloads, secrets, or chunk content. - Run locally with deterministic providers and in-memory retrieval by default, with a low-resource, offline-first baseline.
- Keep FastEmbed, Qdrant, Ollama, Webhook indexing, and public chunk retrieval as explicit opt-ins rather than default behavior.
Memos (Go + React)
├─ Memo data, users, and permissions: source of truth
└─ explicit Webhook integration
│
▼
FastAPI AI Service
├─ AI-derived SQLite state only
├─ templates, insights, and optional index state
└─ provider/vector-store adapters
│
▼
Memo detail view
├─ AI insights review
└─ in-memory Context Pack generation and copy
See docs/structure.md for repository and runtime boundaries, and docs/api.md for API contracts.
Prerequisite: Docker Desktop with Docker Compose.
git clone https://github.com/ToYOhin/devmemo-ai.git
Set-Location devmemo-ai
docker compose config
docker compose up -d --buildAfter startup:
- Memos: http://localhost:5230
- AI Service health: http://localhost:8000/health
The published stable image is also available directly:
docker pull ghcr.io/toyohin/devmemo-ai:stableThe stable image publishes linux/amd64, linux/arm64, and linux/arm/v7
manifests. Download native executables from GitHub
Releases.
The default configuration is intentionally lightweight and conservative:
AI_INDEX_ON_WEBHOOK=false
AI_INDEX_MODE=memo
AI_VECTOR_STORE=memory
AI_PUBLIC_CHUNK_RETRIEVAL=false
- Default Compose does not allow private-network Webhook targets.
- The default stack runs only Memos and the AI Service with modest CPU budgets.
- Qdrant and Ollama require explicit Compose profiles.
- Public chunk retrieval remains disabled. It requires a real trusted gateway, Memos visibility mapping, and a verified disable-and-rollback path.
For a controlled local Docker development topology where a Memos Webhook must
target ai-service, use the docker-compose.local-webhook.yml override
documented in README_AI.md. Do not use that override for a
public or multi-user deployment.
README_AI.md documents deployment configuration, optional AI adapters, and the local Webhook override. Contributions are welcome through CONTRIBUTING.md.
- Setup and usage support: SUPPORT.md.
- Bugs and feature requests: use this repository's GitHub issue forms.
- Security reports: follow SECURITY.md; do not disclose vulnerabilities in public issues.
- Governance, maintainer responsibilities, and release expectations: GOVERNANCE.md.
- For Memos behavior unchanged by this project, use the upstream Memos support and issue channels.
DevMemo AI is distributed under the MIT License. It contains downstream changes based on Memos; upstream attribution and licensing details are preserved in NOTICE and UPSTREAM.md.