Skip to content
Docs

Overview

AgentWatch is Iterate.ai’s AI governance gateway. It sits between an organization’s applications, agents, and employees on one side and every LLM provider on the other — so that every AI interaction can be routed, inspected, costed, controlled, and audited from one place. It answers the questions enterprises cannot currently answer about their own AI usage: who is using what, with which data, at what cost, and against which policy.

Who is AgentWatch for?

Section titled “Who is AgentWatch for?”

Security and compliance teams that need to know what leaves the organization in a prompt; FinOps and IT leaders who need AI spend attributed and capped rather than discovered on an invoice; platform teams standardizing dozens of applications onto one governed AI endpoint; and CISOs dealing with shadow AI — employees using public AI tools with company data on managed devices.

What does AgentWatch do?

Section titled “What does AgentWatch do?”
  • LLM gateway & routing — one OpenAI-compatible endpoint in front of 15 providers and 300+ pre-configured models, with cross-protocol translation, failover, and caching.
  • Governance & cost control — budgets tracked at four tiers, with employee and team budgets enforced at the gateway, not just alerted on, rate limits on every dimension, per-employee attribution, and multi-tenant isolation.
  • Data protection & guardrails — inline DLP for PII, PHI, financial data, and secrets; prompt-injection and content guardrails; and a web application firewall — all applied before a prompt reaches the provider.
  • Shadow AI discovery — find and govern unmanaged public AI usage on managed devices, without a corporate proxy.
  • Observability & insights — full request transcripts, OpenTelemetry traces, Prometheus metrics, dashboards, and a plain-English query interface over your own audit data.
  • Agent fleet & insider risk — govern AI agents as a distinct actor class, with risk scoring and adaptive enforcement that can block an agent before it reaches a provider.
  • Compliance frameworks — select GDPR, HIPAA, SOX, PCI-DSS, SOC 2, ISO 27001, NIST, or HITRUST and have the matching controls and retention activate as a bundle.

What makes AgentWatch different from AI observability tools?

Section titled “What makes AgentWatch different from AI observability tools?”

Most tools in this space observe. AgentWatch enforces, inline and before the request leaves:

  • A budget breach rejects the request rather than emitting an alert after the spend has happened.
  • DLP runs before the provider sees the prompt, so sensitive data is blocked or redacted rather than scrubbed from a trace afterwards.
  • Prompt-injection and policy violations are blocked at the gateway, not annotated post hoc.
  • Governance covers what other tools cannot see: unmanaged web-chat usage on employee devices, MCP tool calls, and autonomous agents as their own risk surface.

How do applications adopt it?

Section titled “How do applications adopt it?”

Three paths, in order of friction. Change one environment variable so an existing OpenAI or Anthropic SDK points at AgentWatch — zero code changes, and the usual route for fleet-wide rollout. Keep your existing wire format, since AgentWatch terminates OpenAI, Anthropic, Google Vertex, Azure OpenAI, and Cohere formats natively. Or run AgentWatch as a transparent proxy, where existing applications route through it with no change at all — not even the URL. Every path keeps full DLP, audit, and budget enforcement.

Where does AgentWatch run?

Section titled “Where does AgentWatch run?”

Self-hosted on your infrastructure (on-premises or your own cloud), as Iterate.ai-hosted SaaS, or hybrid — plus fully air-gapped with no outbound dependency. It is CPU-only software with no GPU requirement, sized from a single container for up to ~300 users to a highly available multi-service deployment for 1,000+. See Deployment.

How does AgentWatch relate to the rest of the Iterate.ai suite?

Section titled “How does AgentWatch relate to the rest of the Iterate.ai suite?”

AgentWatch is the governance layer for everything else. It fronts model traffic from Interplay workflows and Generate assistants, and it governs private models served by Lifeboat on the same terms as commercial providers — one policy, one audit trail, whether the model is yours or OpenAI’s.

Where do I start?

Section titled “Where do I start?”

Pick the row that matches what you are trying to do — each is a complete path, not a prerequisite for the next.

I want to… Start here
Stand up AgentWatch for the first time Install & Setup Guide
Run it on Kubernetes Amazon EKS Install Guide
Point an existing app at it LLM gateway and routing
Call it on behalf of many users from one backend Delegated identity
See what staff paste into ChatGPT and Claude Shadow AI discovery → Endpoint enrollment
Stop sensitive data leaving Data protection and guardrails
Cap and attribute spend Governance and cost control
Connect our identity provider Single sign-on and directory sync
Automate all of it API and automation
Read the whole thing User Guide · Enterprise Guide

The five-minute version

Section titled “The five-minute version”
Terminal window
# 1. is it up, and which version?
curl -s https://<your-agentwatch-host>/health


# 2. which models can this key use?
curl -s https://<your-agentwatch-host>/api/models/available -H "Authorization: Bearer ak_…"


# 3. send one governed request
curl -s -X POST https://<your-agentwatch-host>/v1/chat/completions \
  -H "Authorization: Bearer ak_…" -H 'Content-Type: application/json' \
  -d '{"model":"<one from step 2>","messages":[{"role":"user","content":"hello"}]}'

Then open Usage Insights. That request is now a row with the actor, model, tokens, cost, latency and policy outcome — which is the whole product in one screen.

Sources and references

Section titled “Sources and references”