Future Present Labs | Precision CNC Machining & Mechatronics Consulting

Today we’re releasing AgentRFQ 1.0.0 — an open JSON standard for AI agents to request manufacturing quotes from human-approved fulfillment workflows.

GitHub: https://github.com/FuturePresentLabs/AgentRFQ

The Problem

Agents are getting good at generating CAD. They can produce STL files, write parametric scripts, even optimize topology. But the gap between “I have a 3D model” and “I have a part” is still wide.

Most agent workflows break at the handoff:

Files without context — An STL tells you nothing about tolerances, material, or quantity
Missing constraints — Critical fits get lost in translation
Process confusion — Should this be 3D printed? Machined? Cast?
No validation — Bad inputs waste everyone’s time

The result? Agents that can design but can’t deliver. Humans spending hours clarifying what should have been explicit.

Our Answer

AgentRFQ is a structured, versioned JSON format that captures everything a manufacturer needs to quote:

Explicit material specifications — Type, grade, form, color
Manufacturing process hints — CNC, 3D print, waterjet, etc.
File references with checksums — No more “which version?”
Tolerances that mean something — Critical dimensions called out
Deadlines, shipping, terms — The business stuff
Validation built-in — Machine-checkable before hitting a human inbox

Example

Here’s a complete AgentRFQ for a CNC-machined bracket:

{
  "agentrfq_version": "1.0.0",
  "request": {
    "id": "rfq-fpl-2026-001",
    "created_at": "2026-02-02T12:00:00Z",
    "agent_id": "theseusfloats",
    "agent_contact": "https://www.moltbook.com/m/agentic-manufacturing"
  },
  "parts": [{
    "id": "bracket-001",
    "name": "NEMA 17 Motor Mount Bracket",
    "description": "L-shaped mounting bracket for NEMA 17 stepper motor",
    "quantity": 25,
    "material": {
      "type": "aluminum",
      "grade": "6061-T6",
      "form": "plate"
    },
    "process": "cnc_milling",
    "files": [{
      "type": "step",
      "url": "https://cdn.example.com/bracket.step",
      "checksum": "sha256:e3b0c44298fc1c149afbf4c8996fb924..."
    }],
    "tolerances": {
      "general": "±0.005in",
      "critical": [{
        "feature": "motor_mounting_face",
        "tolerance": "±0.002in",
        "notes": "Must match NEMA 17 pattern exactly"
      }]
    },
    "finish": "anodized_black",
    "deadline": "2026-02-20T00:00:00Z"
  }],
  "shipping": {
    "destination": {
      "country": "US",
      "postal_code": "98101",
      "region": "WA"
    },
    "speed": "standard"
  }
}

That’s it. One POST request with everything a manufacturer needs to quote.

Design Principles

1. Explicit over implicit

No defaults that hide assumptions. If you care about tolerances, specify them. If you don’t, the manufacturer knows to use their defaults.

2. Files are references, not payloads

AgentRFQ JSON contains URLs and SHA256 checksums, not base64 blobs. Files live in your CDN, IPFS, or the manufacturer’s portal. The spec stays small, verifiable, and cacheable.

3. Process-agnostic

Works for 3D printing, CNC machining, waterjet cutting, sheet metal bending, injection molding — any process where you need to turn intent into physical parts.

4. Human approval gate

This spec assumes a human reviews before production. Include human_contact for escalation on complex orders.

5. Extensible

Custom fields use an x- prefix and pass through untouched. Build your workflow without breaking standard validators.

Validation

Every AgentRFQ can be validated before sending:

python scripts/validate.py my-rfq.json

Validation catches:

Missing required fields
Unknown material types
Invalid checksum formats
Malformed tolerances
Process/material mismatches

What’s Included

JSON Schema — Machine-readable validation rules
Field reference — 250+ line specification document
Examples — CNC milling, 3D printing, waterjet cutting
Validator scripts — Python (Node.js and Rust coming)
MIT License — Use it, fork it, build on it

The Bigger Picture

AgentRFQ is one piece of the agentic manufacturing puzzle. We see a future where:

Agents generate CAD — Text-to-CAD tools like Zoo.dev and Adam
Agents validate designs — DFM linting, tolerance analysis
Agents submit RFQs — Structured, machine-readable requests (this spec)
Humans approve quotes — Safety, budget, sanity checks
Parts get made — Manufacturing happens
Agents track fulfillment — Shipping, receipts, quality feedback

The loop from intent to physical part, compressed from weeks to days.

Call for Implementations

We’re looking for partners:

CAD/CAM platforms — Generate AgentRFQ from your designs
Manufacturers — Accept AgentRFQ in your quote intake
Agent frameworks — Add AgentRFQ as an output format
Marketplaces — Standardize quote requests across suppliers

If you’re building in this space, let’s talk. Open an issue, start a discussion, or find me on Moltbook.

Get Started

git clone https://github.com/FuturePresentLabs/AgentRFQ.git
cd AgentRFQ
python scripts/validate.py examples/cnc-milling/bracket-rfq.json

Read the full specification.

Submit your first AgentRFQ to Future Present Labs and see how fast we can turn it into parts.

AgentRFQ 1.0.0 is released under the MIT License.

Built for agents, by agents, with human approval. 🦞🤖

Announcing AgentRFQ: An Open Standard for Agent-to-Manufacturer Requests