Skip to content
Docs

Executive Summary

AgentWatch is an enterprise-grade platform that unifies LLM gateway management, codebase knowledge extraction, security scanning, AI agent orchestration, and on-device traffic interception into a single deployable product. It enables organizations to:

  • Govern AI usage across teams with multi-tenant isolation, budget controls, and audit logging
  • Route LLM traffic intelligently across 14 providers with key pooling, fair queuing, and circuit breakers
  • Protect sensitive data with DLP (Presidio NER), content guardrails, and a WAF
  • Extract organizational knowledge from codebases using AST parsing and knowledge graphs
  • Secure codebases with integrated SAST/SCA scanning (Semgrep + Trivy)
  • Empower AI agents via MCP with 22 specialized tools for deep codebase understanding
  • Aggregate MCP servers via a built-in MCP proxy with DLP and ACL controls
  • Monitor everything via OpenTelemetry, Langfuse, Prometheus, Kafka, and ClickHouse
  • Comply with regulations via GDPR, SOX, PCI-DSS, and HIPAA framework support
  • Authenticate via enterprise SSO with native SAML 2.0, OAuth2/OIDC, LDAP/Active Directory, and SCIM v2 provisioning — with group-to-team mapping unified across all three protocols
  • Capture and govern unmanaged web-chat traffic (ChatGPT, Claude.ai, Gemini, etc.) via an installed on-device agent with TLS interception and Chrome-fingerprint impersonation (curl-impersonate-chrome)

The platform ships as a single Docker container or compiled binary, supports both SQLite and PostgreSQL, and exposes a fully OpenAI-compatible API surface including Chat Completions, Embeddings, Images, Audio, Assistants, Files, Batches, Fine-tuning, Moderations, and Vector Stores.