openbridge-build

openbridge-build — author a new pool producer

This is the library reference for authors. If you're driving an existing pool as a worker, load the openbridge skill, not this one.

What you write

A normal async def main(): that calls await bridge.ask(...) to publish work into a pool. Each ask() is one work item, served by one worker. Concurrent ask()s fan out across workers in parallel.

import asyncio
# Install OpenBridge first:  `pip install -e .` from the OpenBridge repo root
from openbridge import Bridge

bridge = Bridge(name="my-producer", pool="my-pool")
# name = this producer's unique ID (for `openbridge list`, history)
# pool = the work-distribution group workers consume from

async def process(item):
    result = await bridge.ask(
        item_id=item["id"],
        prompt=render_prompt(item),
        template={"summary": "", "scope": ""},
        validate=lambda d: "summary required" if not d.get("summary") else None,
    )
    if result.skipped:
        ...
    else:
        apply(item, result.data)

async def main():
    items = discover_items()
    progress = bridge.checkpoint("progress", default={"done": []})
    pending = [i for i in items if i["id"] not in progress["done"]]
    # Fan out across the worker pool — N items processed in parallel
    sem = asyncio.Semaphore(5)
    async def gated(item):
        async with sem:
            await process(item)
            progress["done"].append(item["id"])
            bridge.save("progress", progress)
    await asyncio.gather(*(gated(i) for i in pending))

if __name__ == "__main__":
    bridge.serve(main())

The runtime handles: scratch submission files per work_id, Redis queue + claim + result lists, validation re-prompt loop, producer/worker registration in pool sets, history.jsonl, state.json.

API

Bridge(name, pool, *, redis_url=None, workdir=None)

• name — this producer's unique identity. Used in openbridge list, history, state.json. Multiple producers with different names can publish into the same pool concurrently.
• pool — the work pool to publish into. Workers consume from this pool. Pool names namespace everything in Redis — two projects with different pool names cannot collide.
• redis_url — defaults to $REDIS_URL or redis://localhost:6379/0. Auto-bootstraps a Docker container if no Redis is up (opt out with OPENBRIDGE_NO_DOCKER=1).
• workdir — defaults to <cwd>/.<name>/. Each producer has its own workdir for state.json / history.jsonl / scratch submission files.