jaros 0.3.2

A zero-infrastructure runtime that makes agent systems reproducible, testable, and capability-safe — a durable, replayable state machine orchestrating AI agents as lightweight threads.

Jaros

A zero-infrastructure runtime that makes agent systems reproducible, testable, and capability-safe by construction — a durable, replayable state machine that orchestrates AI agents as lightweight computing threads, not bloated microservices.

Jaros is the runtime you reach for the day your agent leaves the demo — when non-determinism has made it impossible to reproduce, and ambient power has made it unsafe to ship. It delivers that without a server, a database, or a broker: just files and threads.

It works by decoupling non-deterministic AI reasoning from deterministic system execution. The LLM is an interchangeable application that may only propose inert, serializable Decision data; a deterministic execution plane decides whether and how each decision runs — and may reject it. This is the system's Prime Directive; every part of the codebase exists to serve it.

---

What sets Jaros apart

Most agent frameworks let the model drive: a tool call is a side effect. Jaros inverts it — the model writes recommendations on slips of paper; a deterministic clerk decides what actually happens. Four properties fall out of that design:

• 🐝 Reproducible & accountable swarms — every accepted Decision is recorded in one ordered, hash-chained log tagged with its source agent, so replaying re-executes the whole hive to byte-identical state with zero model calls and attributes any failure to the exact agent and decision.
• 🔁 Reproducible by replay — the only non-determinism is the model's output, recorded before any effect; replaying the log reconstructs the run to byte-identical state with no model call. Crash recovery is just replay.
• 🔒 Capability-safe by construction — agents hold only the scoped handles the harness grants; a bug or bad decision cannot reach what it was never given, and every mediated action is audited.
• 📦 Zero-infrastructure — no server, no database, no broker. The control plane is the file system; agents are threads in one process. A check_zero_infra guardrail fails the build if any code even imports a DB driver or broker.

One command replays a hive and names the culprit — and the agents really let the model decide (accept → DONE, reject → FAILED), yet replay reconstructs whatever the model chose with zero model calls:

→ The full story, with the model-driven decisions, the graduation-layer comparison, and the honest "is / is not": docs/why-jaros.md.

---

Quickstart

Install from PyPI and scaffold a ready-to-run node in two commands:

pip install jaros
jaros init --with-examples   # scaffolds ./.jaros-data with bundled example agents, tools, evals, schedules

jaros init --with-examples drops a library of example agents and tools straight into the data dir, so the daemon — and the console — have something to run immediately. Boot the node, then drive it from another shell; work enters only through the shared file system:

# boot the node (the OS) + the web console — discovers ./.jaros-data by default
jaros serve
#   Jaros node up.
#     data dir : .jaros-data
#     model    : default
#     console  : http://localhost:5500  (starting…)
#     Ctrl-C to stop. The log below shows events as they happen.

jaros serve brings up the web console by default (--no-console to skip) and stays quiet after the banner — it logs only meaningful events (a job completing or failing, a schedule firing), not a per-tick heartbeat.

# from another terminal: submit work + watch results, all over the shared FS
jaros submit system-health             # a bundled example agent
jaros submit advance --input '{}'      # the built-in agent
jaros watch                            # change-only: reprints status when it changes, one line per new result

Then the payoff — reconstruct the entire run from the recorded decisions, with no model call:

jaros replay
#   replayed 3 recorded decisions (3 applied) - model calls: 0
#     reconstructed state : DONE
#     byte-identical      : yes
#   reproducible: the recorded decisions reconstruct the run exactly, with no model call.

The whole loop from the CLI — submit work, check status, replay it byte-identically, and run the eval suite (real output, nothing faked):

Every command discovers the data dir automatically (./.jaros-data, or $JAROS_DATA_DIR, or --data-dir DIR to override). For the full day-one-to-production path (first agent → schedule → eval → replay → console → distributed Docker), see docs/getting-started.md.

Hacking on Jaros itself? Clone the repo and pip install -e ".[dev]" instead — every command works the same against a checkout, and pytest runs the suite plus the architecture guardrails.

Want to build your own agents? Point your coding agent (e.g. Claude Code) at agent-kit/, tell it to read what's there, and it learns the whole system and writes + verifies new Jaros agents for you. See Build an agent.

---

Web console

A TypeScript + React administrative and monitoring interface lives in console/ — submit jobs, install agents and custom tools, watch live status, browse the durable decision log, and replay it to byte-identical state from the browser. It's a host-side companion (a thin file-system bridge + SPA); the Jaros node itself stays serverless.

jaros serve starts it for you and prints the URL — open http://localhost:5500. The Overview is a glanceable NOC view with a live get-started checklist; every other page (State Machine, Reproducibility, Harness, Jobs, Agents, Schedules, Evaluations) introspects the real runtime over the file system.

Want it on its own (first run, or pointed at a remote node's shared dir)? Install its deps once and run it standalone:

cd console && npm install
JAROS_DATA_DIR=/tmp/jaros-demo npm run dev        # then open http://localhost:5500

The full page gallery and a walkthrough of every page (with pictures) live in docs/console.md and the console README.

---

Build an agent

An agent is a ReasoningBoundary: data in → Decision data out, no side effects, no handles. Drop the module into the shared-FS agents/ folder and the daemon registers it at runtime.

import uuid
from jaros.core import create_decision

KIND = "greeter"  # the agent kind the daemon registers

class GreeterBoundary:
    def __init__(self, llm):
        self._llm = llm

    def decide(self, context) -> list:
        name = context.get("name", "world") if isinstance(context, dict) else "world"
        # Propose an inert decision; the executor (not the agent) acts on it.
        return [create_decision(
            id=f"greet-{uuid.uuid4().hex}",
            source="greeter",
            kind="advance",                       # built-in handler drives the state machine
            payload={"events": ["start", "complete"], "note": f"hello {name}"},
        )]

def build(llm):                                   # agent factory the daemon calls
    return GreeterBoundary(llm)

To bound an agent, restrict its capability grant at spawn time — a role is just a named bundle of capabilities. A custom tool extends what the system can do: drop a class exposing NAME, validate(), and execute() into tools/. See examples/tools/greet_tool.py and the full guide in docs/building-agents.md.

Or let a coding agent build it. Jaros is made to be extended by coding agents. Point yours — Claude Code, Cursor, or similar — at AGENTS.md → agent-kit/ and have it read what's there: the mental model, a skill per artifact, accurate API reference, and runnable templates that pass jaros eval unmodified. Tell it "read agent-kit/ and build me an agent that does X," and it will.

---

Run on Docker

The container is the boundary for the whole Jaros node; agents run as threads inside it — never one container per agent.

docker build -t jaros .
docker run -d --name jaros_os -v ${PWD}/.jaros-data:/data jaros   # one daemon = one node
jaros submit advance --input '{}'                                 # submit from the host, over the shared FS

Because the control plane is files only, scheduling needs no broker: any host-side cron can jaros submit, and several daemons can share one directory — each job is claimed by an atomic inbox → claimed rename (exactly-once in the happy path, at-least-once under failure via lease reclaim). The distributed walkthrough is in docs/getting-started.md.

---

Learn more

• docs/getting-started.md — the full day-one path, first agent → distributed Docker.
• docs/why-jaros.md — what sets Jaros apart, in depth, plus where it fits and what it is not.
• docs/architecture.md — the two planes, the guardrails, the subsystem map, and the project layout.
• docs/building-agents.md — write agents and custom tools by hand.
• docs/console.md — a visual guide to every console page.
• agent-kit/ — hand it to your coding agent to build Jaros agents the way you'd use them.
• .jarify/PRIME-001/intent.md — the Prime Directive every part of the system serves.

Jaros is developed spec-first under .jarify/: the Prime Directive holds the system intent, each EXT-00x spec decomposes one tenet into requirements/design/tasks, and code is traced back to requirements via index.json.