Your agents. Your infrastructure. Your rules.
Give agents access.
Keep control.
Connect AI agents to the tools your team runs on. Put policy, approvals and an audit trail between a request and its impact.
Request access Explore the docs
Self-hosted. One Rust binary. Built for developer workflows.
Evaluate the request before it reaches the tool.
- Identity
- Policy
- Approvals
- Audit
- 28integrations, one gateway
- Self-hostedon your infrastructure
- stdio + HTTPfrom laptop to shared service
- Cross-platformLinux, macOS and Windows
01 / CONNECT YOUR STACK
The tools you use. Ready for your agents.
One server for the tools your agents already need, with bounded output and resumable artifacts for large responses.
- Bitbucket Cloud
- Jira Cloud
- Confluence Cloud
- Zoom
- CircleCI
- Slack
- Postman
- edX / Open edX
- New Relic
- Grafana / Loki
- TeamCity
- SonarQube / SonarCloud
- Splunk
- NinjaOne
- GitHub
- GitLab
- Figma
- App Store Connect
- Google Play
- Cloudflare
- Vercel
- Sentry
- Artifactory
- Snyk
- Mend
- WRDS
- Playwright
- AWS
02 / SET THE BOUNDARIES
Make access intentional.
Controls apply to traffic routed through the gateway, with the relevant settings configured.
Verified identity
HTTP clients sign in through your OpenID Connect provider. Sessions, caches and artifacts are bound to the verified principal.
Default-deny policy
Signed policy bundles are evaluated against the upstream request the gateway is about to send. An optional Cedar engine evaluates the same bundles.
Brokered credentials
Upstream credentials stay in the gateway, from the OS keychain, files, Vault, AWS Secrets Manager or Azure Key Vault, and never appear in tool arguments.
Approvals
Hold a call until a second person approves it, before anything reaches the vendor.
Content inspection
Inspect write arguments, tool results, prompts and completions against the policy's content rules, then deny, hold, redact or label. Detection is probabilistic; request rules remain the enforced boundary.
Model requests
Route model provider requests through the same process. Each request is forwarded unchanged to one configured provider, decided on the policy's model rules, and journaled without prompt or completion content.
Audit trail
A sealed audit journal records every decision as metadata, with verification, reports and SIEM forwarding.
03 / INSPECT THE EVIDENCE
Bring your security team into the conversation.
Claims are backed by named tests and published documents. A framework mapping shows traceability; it is not a certification or a statement of legal compliance. Start with the security overview.
OWASP MCP Top 10 assessment
Each category, the controls that address it and the tests that lock them.
Threat model
Trust boundaries, threats, controls and the residual risks we have not closed.
Governance framework map
Traceability from NIST AI RMF, ISO/IEC 42001 and EU AI Act requirements to controls.
04 / RUN IT YOUR WAY
Start on a laptop. Serve your team.
One developer
Run locally over stdio, with credentials in the OS keychain.
A shared gateway
Serve many users over streamable HTTP with inbound sign-in, per-user credentials and the audit journal. Packaged as a container image, Helm chart and Debian package.
Put your agents to work. On your terms.
Binaries and images are available on request while public distribution is set up.