Fig. 01 / The engine floorCode in / Engines out

Autonomous Systems Foundry

Powerful engines.
Open Source Forever.

Farside Research builds autonomous systems for research. The engines that run that work come off our own foundry floor, and we share them: the code under the AGPL-3.0-only, the measurements with their conditions, and the limits we have not solved yet. Teams that keep their own code closed can take a commercial license.

One codebase. Two licenses.
AGPL-3.0-onlyFree. Use, study, change, share. Changes you offer to others stay open.
CommercialClosed products and hosted services. Priced by company size.
The license texts govern. This is the short version.

From $9.99/mo for solo devs, priced by company sizeCommercial licenses are priced by company size, not by seat, monthly or yearly. One license covers every engine we have released.

See license pricing

Our open source pledge

Open Forever.
In Writing.

We build engines for autonomous systems, and we build them in the open. What we release under the AGPL stays open, and we put that in writing.

You have seen how this goes elsewhere: a project grows a community, then the license changes under it. These are our commitments for StateFrame and every other project we release under the AGPL.

  1. Released means released.

    Every version we publish under the AGPL-3.0-only stays under it. We will not withdraw that grant from anyone who complies with the license.

  2. No license switch later.

    We will not move a project to a source-available or other non-open license for future versions. If it ships open, its future ships open.

  3. The whole engine, not a teaser.

    The open release is the full engine, published with complete source no later than anyone else gets it. A commercial license covers the same code on different terms. We do not hold features back for paying customers. The only early access: security fixes may reach affected users up to 30 days before public release, and our own team tests unreleased builds.

  4. Contributions stay in the open.

    When we ship a version that contains your contribution, it ships under the AGPL. The contributor agreement lets us also offer the commercial license. It is never used to take code out of the open.

  5. Fork without asking.

    You can fork any project and ship your own version under the AGPL, under your own name, as long as you follow the license and keep its notices. That right comes with the license. It is not ours to take back.

  6. Open even if we stop.

    If we ever decide to stop maintaining a project, the last release stays AGPL and anyone can carry it forward. We will say so in the project repository and on this site within 30 days, instead of letting it go quiet.

I hate it when companies pull that bullshit on developers. So we put our promises in writing.

Michael AridaFounder / Lead Engineer, Farside Research

Farside Research LLCApplies to StateFrame and every other project we release under the AGPL. The full text is PLEDGE.md in every project repository.

Off the foundry floor

What we have released.

Every engine ships with its own page: what we measured, under what conditions, and what does not work yet.

  • Open Agent EngineAGPL-3.0-only or commercial

    StateFrame

    Runs agents from a typed state frame instead of a growing chat history. Built and measured on software engineering work, with early live results on document and data tasks. Local by default, checked by your own tests, and every request on the record.

    • Local by default
    • Every request on the record

    stateframe run --repo path/to/your-repo --task "Fix the failing pricing tests"

    What we measured

    Prompts that never grow300requests
    Every request fit the 33,344-token budget across a 300-request run. The largest was 32,159 tokens.
    More agents per GPU4 of 4agents passed
    Running at once on one 48 GB GPU in about 26.4 GB of memory. Each ran about 2 times longer than alone, for about 1.9 times the total throughput.
    Real features on an open model17 of 18hidden tests
    A 7-file feature passed every protected test, built by an open model running on one rented cloud GPU.
    Fix weak spots, not the model24 of 24runs
    Early evidence: one fix in the frame took a recurring finishing failure from 21 of 24 runs correct to 24 of 24.
    Measured with one open model, Qwen3.6-35B-A3B, self-hosted on a single 48 GB GPU (RTX A6000 class) rented in the cloud; the 300-request and 7-file results are from the preview engine. Conditions and what does not work yet are on the StateFrame page.

On the bench

  • Business Autonomy EngineIn development

    Momentum Engine

    Companies stall in the gap between noticing and doing. Momentum closes it with bounded autonomy: on a steady rhythm it reads the signals across a business, makes the small, reversible moves people have already approved, brings anything bigger to a person, and then checks whether the move worked.

    Built to

    1. SenseRead the signals across a business on a steady rhythm, not second by second.
    2. DecideTell a blip from a real moment, and size the move to fit it.
    3. ActMake the small, reversible moves people approved. Anything bigger goes to a person, who is always notified.
    4. LearnFollow every move to its outcome. Playbooks that help earn trust; ones that do not are retired.
  • RedactedClassified

    Omnis

    We can't talk about it yet. You'll want to know when we can.

Do you need one?

Most people never pay, by design.

A short guide to which license fits. It is a summary, not legal advice, and the license texts govern.

  1. Free under the AGPL

    Learning, research, and open work

    Use it, study it, change it, and share it. Personal projects, research, and open-source projects that share their code under the AGPL never pay.

    Example use cases

    • A student or researcher experimenting with it
    • Benchmarking it, or writing and teaching about it
    • An open-source project that ships its code under the AGPL
  2. Free under the AGPL

    Running it unchanged on your own machines

    Running an unmodified release for your own work does not ask you to publish anything. If you change it and let others use it over a network, the AGPL asks you to offer them your source.

    Example use cases

    • A developer running it on their own repositories at work
    • A team running the released version on company machines
    • Trying it out before deciding whether you need a license
  3. Commercial license

    Closed products and closed services

    Ship a closed product that includes it, or let customers use your modified version over a network without offering them its source. One license covers your whole company. Changes you only run yourself can stay private under the AGPL.

    Example use cases

    • A closed-source product or app that ships with it inside
    • A hosted service that runs your modified version for customers
    • A client project you deliver without its source
See license pricing

Free to build in the open. Fair to license in private.

Use every project for free in the open. When your own code needs to stay closed, one license covers your whole company.