Open source
$0 forever
- Full safety engine — nothing crippled
- Blast-radius simulation on risky writes
- Human-gated undo with audit trail
- Deterministic YAML policy engine
- Separate control store & durable audit log
The safety layer between
AI agents and your database
Measure the blast radius.
Undo the damage.
Give agents the keys —
and keep the undo button.
“You’re absolutely right. I should not have deleted the production database.” Never let your AI talk to you like that.
AI agents are writing to production databases, right now.
One statement can erase
years of work
Agents type fast. Postgres does exactly what it's told. Database permissions answer who may touch a table — never how much this statement will change.
An undo button for your database, finally
Not just another rule engine. Rules catch the obvious. Interdict measures the actual damage a statement would do — and can take it back after the fact.
Measure, hold, approve, undo.
With approval kept outside the chat.
blocked
DELETE FROM accounts;
no WHERE — would hit every row. fix suggested, agent retries.
held for approval
DELETE … WHERE last_order < '2019'
blast radius 0 rows — measured live, not guessed
operator
$ interdict approve 8f41
approved from your terminal — the token never enters the chat
executed
DELETE 2,300,000
before-images recorded · undo_id 3811adb4
reverted
revert 3811adb4
revert approved by a human — fully audited
Numbers, not promises
2.6µs
added per statement
(warm hot path)
(warm hot path)
≈0ms
end-to-end overhead
p50 & p99
p50 & p99
0%
dangerous statements missed
(red corpus)
(red corpus)
378
automated tests
in the suite
in the suite
Deterministic checks run in memory in microseconds. Simulation is gated to risky writes and time-boxed. The safe path never waits on an LLM.
Give your agents the database.
Keep the undo button.
Open source. Runs locally as an MCP server your agent connects to today — Claude Code, Codex, Cursor, or anything that speaks MCP.
Free for every developer.
Paid where teams need it.
Team Cloud
$149 /mo per database · early access
- Approve held writes from Slack — one tap
- Shared, tamper-evident audit trail
- Dashboard across every agent and repo
- Team roles & approval policies
- Everything in Open source
Enterprise
Custom
- Self-hosted control plane
- SSO / SAML & fleet-of-agents identity
- Retention policies & compliance exports
- Support SLA & onboarding
- Everything in Team Cloud
Be first in line
Team Cloud is rolling out to a small group of design partners. Leave your email and we'll onboard you personally.