Built in · No install

Trust, safety & observability —
wired in by default.

Every agent you build in AgentFoundry comes pre-wired to TrustModel.ai — replay, cost tracking, guardrails, audit logs, eval harnesses, HITL approvals. No SDK install. No separate setup. The same observability surface every Fortune-500 compliance team asks for, shipped as a built-in capability of your IDE.

Download AgentFoundry Visit TrustModel →
× · recommended pairing for compliance-grade agent shipping
No install · Just works

Six trust primitives ship inside AgentFoundry.

The capabilities that compliance teams used to require six different vendors for — observability, audit, guardrails, evals, approvals, replay — are part of the IDE itself. Every agent you scaffold or fork inherits them.

Replay every run

Every agent invocation — local dev and production — captured as a JSONL event stream. Click any past run to step through tool calls, model responses, cost, latency. Fork to current code to see what changed.

Shipped
$

Cost & token tracking

Per-run, per-tool, per-model — pulled directly from the SDK's reported usage. Roll-up across every agent, every invocation, every team. No metering middleware required.

Shipped

Visual guardrails

Data-access policies, allowed-actions whitelists, tone enforcement, escalation rules — set once in AgentFoundry's Guardrails panel, sent as system prompt on every invocation. Domain experts configure them, not engineers.

Shipped

Eval-first dev loop

JSON eval suites with six assertion types. Run-all against every model. Pass/fail diff vs your last run, so you catch regressions before the customer does. Snapshot any successful run as an eval in one click.

Shipped

Human-in-the-loop approvals

When an agent emits a requires_approval tool call, it pauses and renders an Approve/Deny widget in the conversation. Risky actions wait for the human signal. Compliance audit log captures who approved and when.

Shipped

Audit trail · per-agent provenance

Every published agent gets a manifest with built-by, runs-at, data-access scopes, prompts, evals, tool signatures. The provenance follows the agent — fork it, modify it, the chain stays intact.

Shipped
No-install advantage

What every other agent IDE asks you to install separately, ships with AgentFoundry.

Stack the integrations Cursor and competitors point to as "supported via SDK install" — when an enterprise compliance team adds them up, the math gets ugly fast.

Capability AgentFoundry Cursor / Windsurf VS Code + extensions
Replay any production run + Fork to current code Built in No No
Real cost / token / latency dashboard per agent Built in SDK install Helicone / Langfuse / Phoenix
Guardrails as system prompt, set visually Built in No No
Eval suite with pass/fail diff vs last run Built in No Promptfoo / Inspect
Human-in-the-loop approvals as a primitive Built in No No
Per-agent provenance manifest + audit trail Built in No No
Total separate vendor installs needed 0 3–5 5–8
Continuous monitoring across your fleet

Sign in with AgentFoundry. Land in TrustModel Console.

Local dev surfaces what's happening in this agent. The TrustModel Control Plane shows what's happening across every agent your company has deployed.

Coming · Q2 2026

Build in AgentFoundry → monitor in TrustModel — without a second login.

Click Open Control Plane in AgentFoundry's sidebar, hand off to TrustModel via SSO, see real-time alerts on safety violations, population-level analytics, compliance reports (SOC 2, ISO 27001, HIPAA), and adversarial eval coverage across your entire agent portfolio.

One identity. One billing seat. Every guardrail and eval you set in AgentFoundry shows up in TrustModel Console immediately — no sync step, no API config, no JSON to copy-paste.

Visit TrustModel.ai →
1 Build & test agent in AgentFoundry IDE
2 Click Open Control Plane →
3 SSO handoff to TrustModel
(coming Q2 2026)
4 Continuous monitoring + alerts
across all your deployed agents

Ship safer agents without another vendor install.

Download AgentFoundry. Build your first agent in 60 seconds. Trust + safety + observability are already wired in.