IntaGrin

🎯

Declarative, not programmatic

Routing, guardrails, and memory live in ai.yaml, not hand-wired Python graph code or agent/crew objects.

🛡️

Governance built in, not bolted on

Circuit breakers with an honest bounded/unbounded cost accounting, per-call human-in-the-loop approval, content-provenance guardrails, and sandboxed code execution.

✅

Static verification before runtime

inta verify catches routing cycles, unbounded cost paths, and misconfigured guardrails before you ever run the agent.

📖

Honesty over marketing

Every tool in this repo says what it doesn't cover, not just what it does — a claim only ships if the code actually backs it.

The whole idea, in under a minute

No hand-wired routing, no loops nobody planned, no token bill nobody bounded. Declare the swarm instead: one command turns it into a running graph, and every agent loads only the tools it actually needs — smaller prompt, smaller bill. Same commands shown throughout this page, just faster to watch than to read.

Spec-driven, not vibe-coded

Describe what you want in plain English — blueprint.md — and let inta compile turn it into a validated ai.yaml. Re-running it against an updated blueprint diffs against what you already have and merges instead of overwriting, self-healing against schema and router-condition errors before anything ever touches disk.

blueprint.md

markdown

# Vision
A support agent that triages tickets, hands
billing to a specialist, and asks a human
before issuing any refund.

terminal

bash

$ inta compile blueprint.md

✓ Schema + router-condition validated
✓ Scaffolded prompts/triage.jinja2
✓ inta verify: acyclic, cost-bounded

Successfully compiled blueprint into ai.yaml

See everything. Approve anything.

A live, drag-and-drop graph of your actual ai.yaml — not a diagram you drew once and forgot to update. Watch handoffs happen in real time, review exactly which arguments a gated tool call wants to run with before you approve it — including a generated image, not just a wall of JSON — and push a visual edit straight back to ai.yaml on disk. inta monitor — nothing to deploy, nothing to sign up for.

IntaGrinA declarative framework for building multi-agent LLM systems in YAML