Agent Forge Core

A runtime for multi-agent LLM apps that you can test offline and operate with confidence. $297, one-time. Python source, MIT with a no-resale-as-template clause.

Orchestration

Supervisor/worker graph with a dedicated tool-gate node. Runs on LangGraph if installed, or on a built-in executor.

Guardrails

Injection scan, PII redaction (Luhn/IBAN validated), per-tenant budgets, default-deny tool allowlists, human approval.

Operate

Multi-tenant FastAPI with SSE, hashed API keys, OpenTelemetry spans, per-run cost ledger, checkpointed resume.

Verify

41 offline pytest tests, golden YAML evals with a judge rubric, CI workflow included.

Deploy

Dockerfile, compose (API + Postgres + collector), Kubernetes manifests, Terraform skeleton to adapt.

Guides

ARCHITECTURE, GUARDRAILS, EVALS, RUNBOOK, plus four runnable examples.

Quickstart

pip install -e ".[dev]"
pytest
python examples/quickstart.py
agent-forge evals evals/golden.yaml --rubric evals/judge_rubric.yaml

Honest limits

The injection detector is a heuristic first layer, not a guarantee.
Price table values are placeholders; set your provider's prices.
Deploy files are starting points, not a turnkey production stack.
Offline evals prove the plumbing, not the quality of your real model.

*MIT with a clause forbidding resale of the source as a template/boilerplate/course. See LICENSE. Try the interactive guardrail demo.