01 / AUTHORIZE
Access by role, down to the column
Each agent gets its own identity and key. Source roles allow or deny actions per table and column, and anything a role does not allow is refused. Policies only narrow access; they never grant it.
CD-01 · DATA ACCESS · RELEASE CANDIDATE
InterLock
Governed access to production data for AI agents.
A proxy between AI agents and your databases, files and APIs. Every request is authorized by role and policy, sensitive fields are redacted, risky writes wait for a person, and everything is recorded. Your agents and your data stores do not change.
→ agent analyst-claude via MCP query SELECT name, email FROM customers authenticate ........ ok source role ......... allow sales.customers policy .............. allow write risk .......... read only redact .............. email → [REDACTED:EMAIL] audit ............... recorded ← rows returned, redacted
/why
Agent platforms decide which tools an agent may call. Nothing decides which rows those tools may read.
02 / REDACT
Redaction before the agent sees it
Emails, government IDs and other personal data in results are detected and masked in the response, before it leaves your network.
03 / APPROVE
Risky writes wait for a person
Every write is classified by risk before it runs. Updates, unbounded deletes and schema changes queue for a reviewer, with an optional Slack notification.
04 / AUDIT
Every request on the record
Allowed or denied, each request is logged with the identity, the source, the decision and what was redacted. The log exports as CSV.
/how-it-works
One address between your agents and your data.
AGENTS AND CLIENTSClaude CodeCodexGeminiMCP clientspsqlHTTP
INTERLOCK
MCP · Postgres wire · HTTP
Runs inside your environment
DATA SOURCESPostgreSQLMySQLAmazon S3SlackGitHubREST APIs
STEP 1 / DEPLOY
Docker Compose on one machine, or Helm on Kubernetes. Signed image and chart.
STEP 2 / CONFIGURE
Register sources, write roles and policies, and issue one key per agent.
STEP 3 / CONNECT
Point the agent at InterLock. No change to the agent's code.
$ git clone https://github.com/ContextData/interlock $ cd interlock && docker compose up -d # console at http://localhost:9090 - sign in as admin / admin, then set a new password
Claude Code, OpenAI and Gemini have been tested end to end, and any agent that speaks MCP can connect. See every connector and how far it is certified.
/status
What works today, with evidence.
Infrastructure in the path of production data has to be trusted, so every capability carries a status and the evidence behind it. If something is not ready, the feature status page says so.
LICENCE
Apache 2.0
LATEST RELEASE
CLOUD CERTIFIED
DigitalOcean, AWS
ARTIFACTS
Signed image and chart