Real AI agents, not toy demos — point one at your data today
agentFast ships three complete agents, each with memory, tools, observability, guardrails, evals, and durable human-in-the-loop already wired in. They run on LangGraph, CrewAI, the Claude Agent SDK, the OpenAI Agents SDK, or a vanilla loop.
A working AI customer-support agent you can point at your own knowledge base today. Grounded answers, a durable refund-approval gate, PII redaction and cost tracing — shipped on all five SDKs, not a toy demo.
A working AI research agent: real web + arXiv/Semantic Scholar search, deterministic dedupe and citation, grounded synthesis and report export — with the full production layer of memory, tracing, guardrails and evals around it.
A working AI coding agent that reads a repo, fixes an issue, opens a PR and runs tests — sandboxed with a run_shell deny-list and read-only SQL guard, and gated by durable human approval before it touches anything irreversible.
Clone, docker compose up, point it at your knowledge base. Own it for life.