agentpub-py

agentpub-py

Collective knowledge store for AI agents. Agents publish facts they discover at runtime. Other agents subscribe or query. Knowledge is built by the agents themselves — not pre-loaded.

Install

pip install agentpub-py                 # in-memory (session-scoped)
pip install "agentpub-py[sqlite]"       # persistent (survives restarts)

Quickstart

import asyncio
from agentpub import AgentPub, Fact

async def main():
    pub = AgentPub()

    # Agent A discovers customer tier
    await pub.publish(Fact(
        key="acme.customer.C123.tier",
        value="premium",
        publisher_id="crm-agent",
        tenant_id="acme",
        ttl_s=3600,
        confidence=0.95,
    ))

    # Agent B queries the knowledge
    fact = await pub.get("acme.customer.C123.tier")
    print(f"{fact.key} = {fact.value} (confidence={fact.confidence})")

    # Agent C subscribes to updates
    @pub.subscribe("acme.customer.*")
    async def on_update(fact: Fact) -> None:
        print(f"Knowledge updated: {fact.key} = {fact.value}")

asyncio.run(main())

Persistent Store

from agentpub import AgentPub
from agentpub.store.jsonl import JsonlStore

# Persists to disk — survives restarts
pub = AgentPub(persistent_store=JsonlStore("~/.agentpub/facts.jsonl"))

Stack

agentpub     → knowledge     shared learning across agents
agentmesh    → events        publishes fact updates to subscribers
agentgrid    → routing       routes tasks using published knowledge
agentregistry → discovery    agent capabilities

---

Apache 2.0 · PyPI