Read-only first
Agent-written SQL is treated as a proposal until dopedb parses, classifies, and enforces the safety policy.
Open-source local AI database client
dopedbThe AI database client for coding agents
dopedb is a local-first AI database client: talk to your database through an AI agent while credentials, write approvals, rollback previews, and audit logs stay inside a native macOS app.
Tauri nativeCodex poweredLocal audit trail
Safety boundary
The agent can suggest. The app enforces.
dopedb is built for the uncomfortable middle ground: powerful AI help, production database caution, and a human who still needs the final say.
Credentials stay local
Connections and secrets live in the native app boundary instead of being handed directly to the spawned agent.
Auditable by design
Queries, approvals, previews, and execution results are recorded so every database action has a trail.
Daily flow
Built for database work that needs judgment.
- Connect a local, staging, or production database profile.
- Ask Codex to inspect schemas, draft SQL, or explain a result.
- Let dopedb enforce read/write policy before anything runs.
- Approve writes only after rollback preview and risk classification.
codex -> proposed write
UPDATE customers
SET plan = 'pro'
WHERE id = 1842;
dopedb safety:
classification: write
rows estimated: 1
preview: rollback transaction ready
approval: requiredDownload
Install the latest macOS build from GitHub Releases.
The first public channel is macOS through GitHub Releases. The app checks the signed release feed and enables updates from Settings when a newer version is available.
macOS may show a developer warning.
Until dopedb is notarized with an Apple Developer ID, approve it from System Settings, Privacy & Security, Open Anyway after confirming the file came from GitHub Releases.
Project docs