README
agent-app-template
A collection of best practices and patterns for Python AI/ML agent projects — designed to learn from or use as a starting point.
The repo is a monorepo with three packages:
| Package | Description |
|---|---|
agent-lib/ | LangGraph agent library — reusable graph primitives and Postgres checkpointing |
agent-app/ | LangGraph research-assistant reference implementation — FastAPI + Postgres checkpointing |
Highlights
LangGraph agent (agent-lib / agent-app)
- LangGraph patterns — subgraph, fan-out/fan-in (
SendAPI), ReAct, human-in-the-loop (interrupt), reflection, MCP, guardrails, dead letter, time-travel, token streaming - Postgres checkpointing — every state snapshot persisted; full time-travel and replay via API
- Three-layer guardrails — regex → GLiGuard → LLM topic check on input; GLiGuard PII redaction → LLM grounding check on output
- MCP integration —
fastmcpserver exposes tools;langchain-mcp-adaptersbinds them to the ReAct researcher node - Reliability safeguards — configurable ceilings for reflection loops, ReAct steps, LLM timeouts, retries, and total pipeline supersteps
- SSE streaming — writer-node tokens streamed as Server-Sent Events; keepalive ping prevents proxy idle-timeout
- Langfuse observability — optional tracing; configure via
LANGFUSE_*env vars
Contributors
konrad-woj
2 commits
Similar repos
Languages
Python
100.0%
README
agent-app-template
A collection of best practices and patterns for Python AI/ML agent projects — designed to learn from or use as a starting point.
The repo is a monorepo with three packages:
| Package | Description |
|---|---|
agent-lib/ | LangGraph agent library — reusable graph primitives and Postgres checkpointing |
agent-app/ | LangGraph research-assistant reference implementation — FastAPI + Postgres checkpointing |
Highlights
LangGraph agent (agent-lib / agent-app)
- LangGraph patterns — subgraph, fan-out/fan-in (
SendAPI), ReAct, human-in-the-loop (interrupt), reflection, MCP, guardrails, dead letter, time-travel, token streaming - Postgres checkpointing — every state snapshot persisted; full time-travel and replay via API
- Three-layer guardrails — regex → GLiGuard → LLM topic check on input; GLiGuard PII redaction → LLM grounding check on output
- MCP integration —
fastmcpserver exposes tools;langchain-mcp-adaptersbinds them to the ReAct researcher node - Reliability safeguards — configurable ceilings for reflection loops, ReAct steps, LLM timeouts, retries, and total pipeline supersteps
- SSE streaming — writer-node tokens streamed as Server-Sent Events; keepalive ping prevents proxy idle-timeout
- Langfuse observability — optional tracing; configure via
LANGFUSE_*env vars
Similar repos
Contributors
konrad-woj
2 commits
Languages
Python
100.0%