Glossary
The vocabulary of agentic deployment, defined plainly. New to protocols? See how it all fits together.
- A2A (Agent-to-Agent)
- A protocol for agents to advertise skills and collaborate with one another. FastAgentic can expose any endpoint as an A2A skill (v0.3), enabling multi-agent message passing.
- Adapter
- The thin layer that connects a specific agent framework to FastAgentic. Adapters exist for PydanticAI, LangGraph, CrewAI, and LangChain; custom frameworks implement the Runnable interface.
- Agent
- A program that uses a language model to reason over a task, call tools, and produce a result. In FastAgentic, an agent is whatever your framework (PydanticAI, LangGraph, CrewAI, LangChain) produces.
- Audit log
- An immutable record of every agent decision and request, produced by FastAgentic's governance layer for compliance and debugging.
- Checkpoint
- A saved snapshot of a workflow's progress. FastAgentic's StepTracker writes checkpoints per step so a run can resume after a crash instead of restarting.
- Cost tracking
- Per-run, per-user, and per-tenant accounting of token usage and pricing, with hard budget cut-offs that stop runaway LLM spend before it happens.
- Decorator (@agent_endpoint)
- The single Python decorator that turns an async function into a deployed agent endpoint, emitting REST, MCP, and A2A surfaces with shared schemas.
- MCP (Model Context Protocol)
- A protocol that lets assistants and IDEs (Claude, Cursor, and others) discover and call tools. FastAgentic auto-generates MCP tool, resource, and prompt surfaces from your endpoints.
- PII masking
- Automatic detection and redaction of personally identifiable information as it flows through the runtime, part of FastAgentic's built-in governance.
- RBAC
- Role-based access control. FastAgentic enforces scope-based RBAC on top of OAuth2/OIDC so callers only reach the endpoints they are authorized for.
- REST
- The standard HTTP API surface. FastAgentic exposes agents as REST routes with server-sent-event streaming and WebSocket upgrades.
- run_opaque
- A helper that wraps an existing agent object unchanged, snapshotting inputs and outputs so the run becomes resumable without editing the agent's code.
- Runnable
- FastAgentic's generic interface for any callable agent. Implement it to bring a framework that does not have a first-class adapter.
- StepTracker
- FastAgentic's durability mechanism. It records fine-grained, step-level checkpoints to Redis, Postgres, or S3.
- Streaming (SSE)
- Server-Sent Events. FastAgentic streams tokens, tool calls, and intermediate steps to clients in real time, with WebSocket and MCP event streams also supported.
Need FastAPI, LangGraph, or agent platform expertise?
Neul Labs — the team behind FastAgentic — takes on a limited number of consulting engagements each quarter. We help teams ship agents to production, fix broken LangGraph pipelines, and design governance for multi-tenant LLM platforms.