z4j-bare

The framework-free agent runtime for z4j.

Bootstraps a z4j agent inside any Python process, Celery worker, RQ worker, Dramatiq actor, plain script, custom service, and connects it to z4j over an authenticated WebSocket. Used directly when there's no framework adapter (z4j-django / z4j-flask / z4j-fastapi); used indirectly by those framework adapters too.

Compatibility

Python 3.11+. No framework or engine pinned; pair with whichever engine adapter your worker runs (z4j-celery, z4j-rq, z4j-dramatiq, z4j-huey, z4j-arq, z4j-taskiq) and that adapter carries the engine-version floor.

Full per-adapter matrix at https://docs.z4j.com/reference/compatibility/.

What it ships

• Agent runtime, connect, authenticate, supervise, reconnect with bounded backoff
• Outbound buffer, captured events are written to a bounded local SQLite ring before going out on the wire and survive short brain outages and agent restarts. At the configured count or byte limit, the oldest buffered rows are evicted and the loss is logged.
• Engine signal hooks, wired up by whichever engine adapter you install (z4j-celery, z4j-rq, etc.); the runtime drains them into the buffer
• Schedule inventory, emits a full snapshot at boot, on a periodic timer (default 15 min), and on demand from the brain's Sync now command. Existing schedules show up automatically.
• Command dispatcher, receives operator actions from the brain (retry, cancel, restart, schedule.fire, schedule.resync, etc.) and routes them to the right adapter

Install

pip install z4j-bare

Most users install a framework adapter (z4j-django / z4j-flask / z4j-fastapi) which pulls z4j-bare automatically.

Quick start (framework-free worker)

import os

from z4j_bare import install_agent
from z4j_celery import CeleryEngineAdapter

install_agent(
    engines=[CeleryEngineAdapter(celery_app=app)],
    brain_url="https://brain.example.com",
    token="z4j_agent_...",
    project_id="my-project",
    hmac_secret=os.environ["Z4J_HMAC_SECRET"],
)

Durable native schedule control

from z4j_bare import NativeScheduleControl provides the private SQLite store used by the Huey periodic, arq cron and TaskIQ label-source adapters. Construct it with a stable application-local file path and pass it as control= to the scheduler adapter in the actual native producer. Back up this file: it retains dashboard-created schedules, field overrides, pauses and deletion tombstones. The producer consumes it without requiring a live server connection. See each scheduler's setup guide for startup ordering.

Reliability

• Invalid or missing configuration and unsafe buffer initialization fail the explicit install_agent() call; after successful startup, transport work runs off the host application's request and task paths.
• Engine adapters use bounded in-process event queues and the runtime uses a bounded SQLite buffer. A full engine queue loses an event: the Celery, RQ and Dramatiq adapters evict the oldest queued event, the Huey, arq and TaskIQ adapters drop the new one. Buffer pressure evicts the oldest rows. Both paths log the loss.
• The supervisor retries every failure with bounded, jittered backoff until the agent is stopped. Authentication and protocol-version failures are retried on much longer schedules (up to 10 minutes and 1 hour between attempts) so they do not create reconnect storms.

Documentation

Full docs at docs.z4j.com/frameworks/bare/.

License

Apache-2.0, see LICENSE.

Links

• Homepage: https://z4j.com
• Documentation: https://docs.z4j.com
• PyPI: https://pypi.org/project/z4j-bare/
• Issues: https://github.com/z4jdev/z4j-bare/issues
• Changelog: CHANGELOG.md
• Security: security@z4j.com (see SECURITY.md)