CAIN-42 CAIN StudioDevelopers
Developer documentation
Install, connect, protect, run, inspect, audit. Start with the quickstart; every page is one click from every other.
Last reviewed 31 August 2026
The ecosystem and libraryEvery product on one page, 14 niches, live pages, and in-browser verification.CAIN-42 todayThe state of the technology as of 1 October 2026: what is live, how a decision is made, what is not.QuickstartNothing to a protected agent, in five minutes.Fast pathTrusted agents skip the per-call consensus round: about 70 ms instead of about 600 ms.Use with your existing stackLangChain, OpenAI Agents SDK, CrewAI, MCP: add a decision before every tool call, no SDK needed.CLI referenceAll 14 commands, exit codes, and --json.Python SDKpip install the hosted cainstudio wheel; @cainstudio.guard() and approvals.TypeScriptNo npm package yet: a dependency-free helper for Node 18+, Deno, Bun.MCPPut MCPGate in front of an MCP server.IntegrationsLangChain, LangGraph, CrewAI and the rest -- with honest status.PoliciesDefault deny, dry runs, and policy versions.ActionProofVerify a plan before any of it runs.EvidenceWhat is recorded, and how to search it.Decision record formatOpen specification: verify any CAIN decision offline, without trusting CAIN.SIEM forwardingSend every decision to Splunk, Cortex XSIAM, Datadog, Sentinel, Slack, Teams or any HTTPS endpoint.Single sign-on and SCIMSign in to the console through Okta, Entra ID or any OIDC provider; provision members with SCIM.Credential brokerAgents call third-party APIs through CAIN and never hold the key.Beside your gatewayKeep Portkey, Kong, Cloudflare or your AI security platform; add decisions on tool calls.CAIN TraceReal execution history with complete lifecycle correlation.ConformanceThe suite that can fail, and why a skip is not a pass.TroubleshootingStart with cain doctor.Developer portalThe engineering control centre, and how it is governed.MarketplaceExtensions: permissions, signing, and what install refuses.Free tierWhat free includes, what costs money, and why.BenchmarksMeasured latency and throughput. No modelled figures.ArchitectureRead this when you want to, not before you start.CAIN IdentityVerifiable identity for agents, humans, and services.CAIN ControlAgent, tool, and policy control.CAIN BudgetHard spending limits enforced before every AI action.CAIN GovernanceOrganizational rules, authorities, and approval workflows.CAIN MemoryGoverned persistent memory for autonomous agents.Self-Hosted MCPGateDeploy CAIN's enforcement boundary into your own VPC.CAIN PrivatePrivate preview: an isolated agent environment.CAIN TrajectoryGovern the whole path, not just the next action.CAIN Agent SecurityReal-time threat detection and enforcement for AI agents.CAIN DriftDetect meaningful change in autonomous systems before it becomes an incident.7-Moat ArchitectureThe seven defensibility layers, as built.ChangelogNotable changes, milestones, and releases.