FAFO™ AgentOS

Governed autonomous work.

9 followers

Governed autonomous work.

9 followers

FAFO™ AgentOS is the platform for governed autonomous work. You describe what needs to happen and AgentOS derives both how it runs and how it is governed. The autonomous workforce (FAFO™ Agent Swarm) performs the work, a governance engine gates every step on evidence rather than assertions, and a Postgres work graph is the system of record that answers who did what, when, and why. It runs on FAFO™ Inference Fabric for execution and FAFO™ Memory for grounding.

Payment Required

Launch tags:User Experience•Software Engineering•Artificial Intelligence

Launch Team / Built With

Framer AI AgentsDesign and publish professional sites with AI

Promoted

Maker

📌

I kept hitting the same wall building real software with AI agents. They'd do impressive work, and then I had no idea what they'd actually done, why, or whether I could trust it. The state lived inside the model, so the moment a model changed or a run failed, the whole thing fell apart. No audit trail, no governance, no way to answer "who did what, when, and why." That's the problem FAFO™ AgentOS solves. Instead of prompting an agent and hoping, you describe the work in plain language and AgentOS derives both how it runs and how it's governed: the workforce, the task graph, the contracts, the budgets, the recovery topology, and the evidence each step has to produce. A Postgres work graph is the system of record, and it lives outside the model, so the work survives even as models, providers, and pricing change. Gates pass on evidence, not assertions. The biggest shift in how I built it: I stopped trying to make the agents smarter and started making each job small and externalized. Once authority, state, and evidence live outside the model, every task is small enough that there's nothing to "remember," and the work becomes governable and auditable by construction. That reframe is what turned a pile of agents into an operating system.

Report

2mo ago

A "dry run" or shadow mode would be huge for adoption. Let me point AgentOS at a real workflow and watch it propose a plan and governance checks before any actual execution happens. That way I can build trust in the evidence gates without risking anything, and it would make demos to my team way more convincing.

Report

1mo ago

Maker

@nevzatakal1e4q This is basically how AgentOS already works — the "shadow" phase isn't a mode, it's the front half of the pipeline. Every piece of work starts as a work order that gets authored before anything can execute. Point it at a real system and the authoring pass:

Grounds through FAFO Memory — semantic code search, a dependency graph, and the project's recorded decision history — so the plan maps your actual code, not an LLM's guess about it
Produces the full plan as reviewable artifacts: objective, scoped envelopes, acceptance criteria with explicit close bars, and the part you're asking about, the complete gatekeeper requirements up front: every proof the evidence gate will demand, each with a positive case and a negative control ("inject an unproven field → gate must reject")
Writes it all to markdown + Postgres rows you can read line by line, in a full plan for approval before execution.

Nothing executes at that point...no team, no dispatch, no side effects. You're reading the governance contract the machine will hold itself to. Execution begins only when you approve the plan, and even the execution-time verbs have dry-run forms that print exactly what they would do (would_mutate: false) before you let them.

The deeper trust mechanism: every packet an agent receives embeds its response contract, pre-send checks refuse un-evidenced work at zero cost, and closes are deterministic coverage proofs, the work order closes when every close item is covered by a current PASS verdict.

Report

1mo ago

finally sat down and poked around with fafe agentos and honestly the evidence gated governance angle is what got me, feels like every action has a clear trail instead of just vibes

Report

1mo ago

Forum Threads

p/fafo-agentos

•

2mo ago

We're missing an entire layer of the AI stack

I've spent the last few years building agentic systems. Going into that work, I assumed the hardest problems would be about the models themselves: better reasoning, larger context windows, stronger planning, and richer tool use. That's exactly where most of the industry's attention has been, and the progress has been remarkable.

Over time, though, the models gradually stopped being the thing I thought about every day.

Instead, I found myself spending more time on a different class of problems: the same kinds of problems every operating system, database, or distributed platform eventually must solve. What is this agent actually allowed to do? How do we know if the work is complete? Where does the state live? What happens when something fails halfway through? If the model changes next month, what will survive?

View all

Grounds through FAFO Memory — semantic code search, a dependency graph, and the project's recorded decision history — so the plan maps your actual code, not an LLM's guess about it
Produces the full plan as reviewable artifacts: objective, scoped envelopes, acceptance criteria with explicit close bars, and the part you're asking about, the complete gatekeeper requirements up front: every proof the evidence gate will demand, each with a positive case and a negative control ("inject an unproven field → gate must reject")
Writes it all to markdown + Postgres rows you can read line by line, in a full plan for approval before execution.