The Infrastructure Layer for LangGraph Agents

~200
LOC adapter
50
MCP tools
5
SDKs, one layer
1 cmd
docker compose up

agentFast is the production layer for LangGraph. LangGraph gives you the graph — nodes, edges, state, and genuinely excellent control flow. agentFast gives you everything that has to sit around it before a LangGraph agent can go in front of a paying customer, so you keep writing real LangGraph and stop rebuilding the plumbing.

What agentFast adds to LangGraph

Native interrupt + checkpointer

A native interrupt() for HITL and a real checkpointer wired to a durable Postgres store — a run survives a process crash.

~200-LOC adapter, not a wrapper

It uses LangGraph the way LangGraph wants to be used. Your graph definition stays yours — never hidden behind a grand abstraction.

Compaction, built in

Context compaction is wired straight into the adapter, so long conversations don't blow the window.

50 MCP tools

A registry of guardrailed, traced, rate-limited tools — add any MCP server by config, zero orchestration code.

Full observability

Every run is a step tree with per-step tokens, cost, latency and prompt-cache accounting — plus OTEL and LangSmith export.

Guardrails + evals

PII redaction before the model sees it, rate limits, and evals that fail CI on a regression.

From clone to a running LangGraph agent
  1. 1
    Install with the LangGraph extra

    One pip install pulls the adapter and dev tools.

  2. 2
    Scaffold an agent

    Generates config, a kb/ folder, seeded evals, and a compose file.

  3. 3
    Bring up the stack

    Postgres, Redis, the agent API, dashboard and playground — one command.

  4. 4
    Gate on quality

    agentfast eval exits non-zero if a metric drops below its threshold.

bash
pip install -e ".[langgraph,dev]"
agentfast new support --sdk langgraph
cd support-agent && docker compose up
agentfast eval
LangGraph vs. agentFast on LangGraph
LangGraphagentFast on LangGraph
Graphs, nodes, state machines✓ its core strengthUses LangGraph directly
Checkpointer / interruptPrimitivesWired to a durable Postgres store
Memory (vector + episodic)— bring your own✓ Included
50 MCP-native tools— bring your own✓ Included
Observability + cost tracing— bring your own✓ Included
Guardrails (PII, safety, limits)— bring your own✓ Included
Durable human-in-the-loop + UIPrimitives onlySurvives a crash, exactly-once
Evals in CI— bring your own✓ Included
Working reference agentTemplatesSupportAgent, point it at your KB
The property most LangGraph tutorials skip
Kill the process while a LangGraph run is paused for human approval. Restart it. Approve. The run resumes from the exact step — no replay, no double-execution. It's a test that runs against real Postgres with a real SIGKILL across two OS processes. See how durable execution and durable human-in-the-loop work.

Already on LangGraph? You lose nothing — agentFast supports five SDKs behind one layer, so the orchestrator stays your choice. Compare LangGraph vs CrewAI vs a vanilla loop, read the direct agentFast vs LangGraph breakdown, or weigh the cost of rolling your own.

Stop rebuilding the plumbing.

agentFast is the production layer — memory, 50 tools, observability, evals, guardrails, human-in-the-loop — for LangGraph, CrewAI, the Claude & OpenAI Agent SDKs & Vanilla. Own it for life.

Get agentFast — from $89
Keep reading
The Production Layer Every AI Agent Needs (and Why LangGraph Doesn't Give It to You)
9 min read
LangGraph vs CrewAI vs a Vanilla Loop: Which Should You Actually Use in 2026
11 min read
How to Add Human-in-the-Loop Approval to Any AI Agent (With Durable Execution)
10 min read