Connect identity, policy and action.
Discover what is running, put one check on the execution path, and keep evidence for every decision.
One connected platform
Five stages, one policy. Start anywhere and add the rest as you need it.
- ConnectGateway and SDKs on the execution path.
- IdentityWho is calling, and for whom.
- GovernOne policy, tested before release.
- ObserveDiscovery, change review, coverage.
- RespondInvestigate and act on what is found.
Two enforcement points. One policy.
The gateway protects requests travelling to services. The SDK protects actions inside the agent application. Both use the same AgenticDome policy and runtime services, so the choice is where the check sits, not which rules apply.
| If your agent | Use |
|---|---|
| Calls MCP servers, APIs, GraphQL or other agents | Gateway |
| Takes consequential actions inside application code | SDK or runtime plugin |
| Runs Docker Agent shell, file or local tools | Docker mandatory native guard + assigned sidecar |
| Does both | Gateway and SDK |
| Might go around the check | Add Runtime Grid discovery and governed egress |
Reuse the identity you already have.
Use tokens from your identity provider, such as Entra ID or Okta, or identities issued by AgenticDome. Verified caller identity and available delegated authority inform connected gateway and supported SDK decisions. Deployments, provider trust, credential rotation and revocation are managed, and history is kept.
From discovery to outcome, six steps.
- 1 · DiscoverFind the agent and its tools.
- 2 · AttestVerify who is calling.
- 3 · AuthoriseDecide against policy.
- 4 · AdmitCheck exact-request authority before the side effect.
- 5 · ExecuteForward or run the admitted request once.
- 6 · VerifyRecord the outcome as evidence.
Outcome evidence is kept distinct by source: reported by the SDK, observed at the gateway, or destination-attested through a configured destination adapter that signs a receipt of the business result. Enforcement applies only to execution paths connected through supported integrations. See the workflow control points →
One action policy, released with care.
Three kinds of evidence, one investigation.
A network connection alone does not show intent or business outcome, so the Grid correlates three layers.
What exists, what is used, who called it, and what changed.
Connect a source and AgenticDome finds agents, MCP servers and models. Each record separates what is configured from what is actually used, and keeps the evidence that found it. Sample data below.
| Name | Found by | Exists | In use | Who called it | Execution path |
|---|---|---|---|---|---|
| refund-agent | SDK, gateway, Runtime Grid | Configured | 4 minutes ago | Workload and 2 users | Protected, verified |
| analytics-mcp | MCP inspection, registry card | 4 tools, 2 new | Yesterday | sales-research-bot | Not governed |
| invoice-extractor | Runtime Grid | Process seen | Provider traffic seen | Workload only | Suspected bypass |
Finding something never grants it access or trust. Importing telemetry gives visibility. Active protection starts only when the execution path is connected.
Discover, connect protection, verify a real action, monitor.
Investigate and act with scope.
Your deployment concierge.
Tell Abi what you need your agents to do. Abi compares gateway, SDK, Docker native guard, identity and runtime options, explains the trade-off, and produces a deployment blueprint for your team to review. Ask it about MCP schema checks and how to verify each boundary. AgenticDome Copilot then guides setup and verification.
See how Abi and Copilot work together