A developer's second brain that passively captures what you do, understands why you did it, and proactively tells you what you need to know.
Developers lose massive amounts of context constantly — switching tasks, coming back after a weekend, debugging something they fixed months ago. DevBrain eliminates this by running as a background daemon on your machine, passively capturing your AI-assisted coding sessions, and building a knowledge graph of your decisions, dead ends, and patterns. It then proactively surfaces relevant context before you need to search for it — morning briefings, "you tried this before" warnings, and semantic search across your entire development history.
Everything stays on your machine. No telemetry. No cloud sync. Your data is yours.
- Passive capture — Automatically ingests AI coding sessions (Claude Code, Cursor) with zero friction. No manual logging.
- Knowledge graph — Builds a graph of decisions, dead ends, patterns, files, and threads. Understands how everything connects.
- Morning briefings — Start your day with a summary of where you left off, what changed overnight, and what to watch out for.
- Dead end warnings — "You tried this approach 3 weeks ago on the auth service. It failed because of X." Saves hours of repeated mistakes.
- Semantic search — Search your entire development history by meaning, not just keywords. Combines vector search, full-text search, and graph expansion.
- Privacy-first — Local-only by default. Automatic secret redaction (API keys, tokens, passwords). Configurable strict mode strips raw content entirely.
- Web dashboard — Timeline view, thread browser, dead end catalog, briefing viewer, and system health — all at
localhost:37800.
# Install
curl -fsSL https://raw.githubusercontent.com/devbrain-tool/devbrain/main/scripts/install.sh | bash
# Start the daemon
devbrain start
# Check that everything is running
devbrain status
# Open the web dashboard
devbrain dashboardAI Sessions ──> Capture Pipeline ──> SQLite + Knowledge Graph
|
CLI / Dashboard <── Intelligence Agents ──> Briefings
DevBrain runs as two binaries:
- Daemon (
devbrain-daemon) — background process hosting the capture pipeline, intelligence agents, storage, and web dashboard. - CLI (
devbrain) — thin HTTP client that talks to the daemon. Commands feel instant because the CLI doesn't initialize storage or agents.
The capture pipeline processes events through five stages: Normalize (schema), Enrich (project/branch/thread), Tag (classify via local LLM), Privacy Filter (redact secrets), and Write (persist to SQLite + index to LanceDB). Intelligence agents then run on schedules or in response to events, building the knowledge graph and generating briefings.
| Command | Description |
|---|---|
devbrain start |
Launch the daemon as a background process |
devbrain stop |
Graceful shutdown |
devbrain status |
Health check, storage stats, agent status, LLM connectivity |
devbrain briefing |
Show today's morning briefing |
devbrain briefing --generate |
Force-regenerate the briefing |
devbrain search <query> |
Semantic + full-text search across all observations |
devbrain search --exact <query> |
Full-text search only (FTS5) |
devbrain why <file> |
Show decisions, dead ends, and patterns for a specific file |
devbrain thread |
Show the current active thread |
devbrain thread list |
List all recent threads |
devbrain thread new <title> |
Manually start a new thread |
devbrain dead-ends |
List dead ends for the current project |
devbrain dead-ends --file <path> |
Dead ends for a specific file |
devbrain related <file> |
Graph traversal: everything connected to a file |
devbrain agents |
List agents, their status, and last run time |
devbrain agents run <name> |
Manually trigger an agent |
devbrain config |
Show current settings |
devbrain config set <key> <value> |
Update a setting |
devbrain export |
Full data export (JSON) |
devbrain purge |
Delete data by project or date |
devbrain rebuild vectors |
Re-embed all observations from SQLite |
devbrain rebuild graph |
Replay the Linker Agent over all observations |
devbrain dashboard |
Open the web dashboard in your browser |
devbrain service install |
Install as a system service for auto-start |
devbrain service uninstall |
Remove the system service |
devbrain update |
Update to the latest release |
devbrain update --check |
Check for available updates |
curl -fsSL https://raw.githubusercontent.com/devbrain-tool/devbrain/main/scripts/install.sh | bashirm https://raw.githubusercontent.com/devbrain-tool/devbrain/main/scripts/install.ps1 | iexDownload the latest release for your platform from GitHub Releases. DevBrain ships as Native AOT binaries — no .NET runtime required.
| Platform | Binary |
|---|---|
| Linux x64 | devbrain-linux-x64.tar.gz |
| Linux ARM64 | devbrain-linux-arm64.tar.gz |
| macOS x64 | devbrain-osx-x64.tar.gz |
| macOS ARM64 | devbrain-osx-arm64.tar.gz |
| Windows x64 | devbrain-win-x64.zip |
| Windows ARM64 | devbrain-win-arm64.zip |
Extract and add the directory to your PATH.
git clone https://github.com/devbrain-tool/devbrain.git
cd devbrain
dotnet build
dotnet run --project src/DevBrain.Api # Start the daemon
dotnet run --project src/DevBrain.Cli # Run CLI commandsDevBrain stores its configuration at ~/.devbrain/settings.toml. A minimal configuration looks like:
[capture]
sources = ["claude-code", "cursor"]
[llm.local]
provider = "ollama"
model = "llama3.2"
embedding_model = "nomic-embed-text"
[llm.cloud]
provider = "anthropic"
# Set ANTHROPIC_API_KEY environment variable
max_daily_requests = 50
[privacy]
mode = "redact" # "redact" or "strict"
[agents]
briefing_time = "07:00"
compression_after_days = 30Run devbrain config to view all current settings, or devbrain config set <key> <value> to change individual values. Most settings are hot-reloaded — no daemon restart required.
For the full configuration reference, see the Configuration section above.
DevBrain is designed with privacy as a core constraint, not an afterthought.
- Local-only — All data stays on your machine in
~/.devbrain/. No telemetry, no phone-home, no cloud sync. - Automatic secret redaction — API keys, tokens, passwords, PEM keys, and
.envvalues are detected and replaced with[REDACTED:type]before storage. Original content is never persisted. - Strict mode — For maximum privacy,
mode = "strict"stores only summaries — no raw content, no file contents, no conversation text. - Cloud LLM boundary — When using cloud LLMs (for briefings), only summaries and metadata are sent. Raw content, file diffs, and secrets never leave your machine.
- File permissions —
~/.devbrain/is locked to your user account on first run (Unix700, Windows user-only ACL). .devbrainignore— Exclude specific files or projects from capture entirely.
| Version | Focus | Highlights |
|---|---|---|
| v1.0 | Core (current) | Capture pipeline, knowledge graph, briefings, dead end detection, semantic search, CLI, web dashboard |
| v1.1 | IDE integration | VS Code extension, JetBrains plugin — inline dead end warnings, contextual search |
| v2.0 | Deeper intelligence | Pattern detection across threads, git adapter (PR/merge context), recurring bug alerts |
| v3.0 | Team sync | Optional encrypted team knowledge sharing, shared dead end catalog, cross-developer patterns |
We welcome contributions. See CONTRIBUTING.md for guidelines on:
- Setting up the development environment
- Running tests (
dotnet test) - Code style and conventions
- Submitting pull requests
If you're looking for a place to start, check issues labeled good first issue.
DevBrain builds on excellent open source software:
- SQLite — storage engine
- LanceDB — vector search
- Ollama — local LLM inference
- ASP.NET Core — daemon host and API
- System.CommandLine — CLI framework
- React + Vite — web dashboard