Agent Phabric Documentation
Welcome to Agent Phabric, the platform that makes managing AI agents simple.
What is Agent Phabric?
Agent Phabric is a comprehensive platform focused on observability for AI agents. Building a single agent is straightforward, but operating many agents in production—while understanding why they act, how they interact, and where they fail—requires deep visibility.
We handle all the "Day 2" operational challenges so you can focus on what matters: building amazing AI agents.
The Problems We Solve
- Observability: How do I see every tool call, prompt, model output, and decision?
- Debugging: How do I reproduce failures and replay conversations deterministically?
- Experimentation: How do I evaluate and compare runs across versions and prompts?
- Security: How do I ensure only authorized users and agents can communicate?
- Cost & Performance: How do I track latency, tokens, and regressions over time?
Observability
Unified Traces
End-to-end traces that stitch prompts, tool calls, webhooks, and downstream services into a single view.
Structured Logging
Opinionated, JSON-first logs with correlation IDs so every event is attributable to a user and a run.
Replay & Time Travel
Deterministically replay conversations and workflows to isolate prompt or tool regressions.
Metrics & Experimentation
Track latency, tokens, and success rates out of the box. Ship with confidence using built-in A/B testing, Blue/Green deploys, and Human-in-the-Loop—all from day one, zero downtime required.
Platform Features
Instrumentation SDKs
Drop-in SDKs for Python/TypeScript automatically capture prompts, tool calls, model IO, and breadcrumbs with zero manual logging.
Agent Discovery & Communication
Built-in service discovery and multiple communication patterns including direct calls and event-based messaging for complex multi-agent workflows.
Performance & Cost Insights
Track latency distributions, token usage, and long-tail failures across agents and versions.
Enterprise Security
Token-based authentication, fine-grained access control, and policy-as-code ensure your agents operate securely at scale.
Framework Agnostic
Agent Phabric works with any AI agent framework or library. Whether you're using popular frameworks or building custom solutions, you can instrument and monitor your agents seamlessly.
Popular Frameworks
- LangChain: Agents, chains, and tools
- CrewAI: Multi-agent collaboration systems
- AutoGen: Conversational AI agents
- Agno: Autonomous agent frameworks
Easy Instrumentation
Already have agents built with your favorite framework? No problem. Wrap key functions with our SDK, set an API key, and gain instant traces, logs, and metrics—no infra changes required.
Security & Compliance
Security is foundational to Agent Phabric. We use enterprise-grade, token-based authentication and provide multiple layers of protection for your AI agents.
Authentication & Authorization
- • JWT-based authentication
- • Role-based access control (RBAC)
- • Team-based permissions
- • Workload identity for agents
Data Protection
- • End-to-end encryption
- • Network isolation
- • Secrets management
- • Audit logging
Ready to Get Started?
Join our early community to contribute ideas, test new features, and shape what comes next.