Skip to main content

​
What is Sentrail?

AI agents write code, create issues, merge pull requests, and update project trackers. Most tools give them full API access — no oversight, no audit trail, no way to stop a bad action before it lands. Sentrail fixes this. Your agents send requests to Sentrail instead of directly to GitHub, Linear, or your MCP server. Sentrail evaluates every action against your policy set, then either:
  • Allows the action and forwards it immediately
  • Blocks it and returns a 403 with a reason
  • Queues it for human approval and returns a 202 with a poll URL
The agent, the tool, and every decision are recorded in an immutable audit log.

​
The approval gateway concept

Agent → Sentrail Gateway → Policy Engine → decision
                                              ↓
                                    allow  block  require_approval
                                      ↓              ↓
                                  Tool API      Approval queue
                                                  → reviewer decides
                                                  → execute or deny
When a request is deferred, the original HTTP call — headers, method, body, target URL — is stored in a deferred_actions row. Once a reviewer approves, Sentrail replays the exact request to the tool API. The agent can poll /gateway-proxy/status/:correlationId for the result.

​
When to use Sentrail

  • You are deploying AI coding agents (Claude Code, Cursor, custom LLM pipelines) that write to production repositories
  • You need an audit trail of every AI-initiated write action
  • You want human-in-the-loop approval for high-risk operations without changing agent code
  • You are working toward SOC 2 or EU AI Act readiness and need demonstrable AI oversight controls

​
Core integrations

IntegrationTransportStatus
GitHubOAuth + REST proxyLive
LinearOAuth + webhookLive
MCP serversJSON-RPC over SSE/HTTPLive
SlackWebhook (notifications)Live
Notion—Coming soon
Internal API—Coming soon
Sentrail proxies GitHub and Linear API calls directly. For MCP servers, Sentrail wraps the tools/call method — the agent connects to mcp-gate instead of the upstream server.