blackboardx

A skeletal blackboard system for Python.

The blackboard architecture came out of HEARSAY-II, a speech understanding system built at Carnegie Mellon in the early 1970s under a DARPA programme. Its difficulty was that a stretch of speech admits several readings, and the knowledge that settles which one is right arrives in unrelated kinds: acoustic, lexical, syntactic, semantic. Which kind will settle a given stretch is not known until that stretch is examined, so the system could not be written as procedures calling one another, because a call fixes what runs next. HEARSAY-II gave its specialists a shared structure to work on instead. Each reads what bears on its own expertise and writes back what it concludes, and none of them calls another.

Later systems kept that arrangement and replaced the knowledge: HASP interpreted sonar rather than speech. H. Penny Nii, surveying blackboard systems in AI Magazine in 1986, named a system skeletal when it supplies the components alone and leaves the knowledge and the control to whoever builds on it.

blackboardx is skeletal in that sense. It supplies the board, which stores what agents write and puts every write in one order, and the control component, which determines who is notified of a change, which writes are admitted, and when the run ends. An application supplies its regions, their opening premise values, the agents the run starts with, an admission rule, a termination predicate, and limits.

The record outlives the run that wrote it. create_model opens a board the store does not hold yet; attach_model opens a run over a board the store already holds, and continues the sequence from where the record ends.

The library also carries both halves of the conversation between a blackboard and agents deployed as their own services: the bodies and operations they share, the piece that answers an agent's request, the piece that sends a notification without making the writer wait, and the client an agent calls with. Your service keeps its own HTTP server, its routes, its authentication, and its database; the library supplies the protocol between them.

An agent reads and writes through AgentBoard, which is the four reads and the three writes without the agent's own name. Control.as_agent returns an AgentBoard for an agent in the same process as the run, and BoardClient is one over HTTP, so an agent body is written once and deployed either way.

What an agent knows is the application's to supply. An agent whose expertise is an algorithm decides in its own code. An agent whose expertise is a language model puts the decision to the model, offering it the board as tools through blackboard.tools, and makes each call the model asks for. The library sends nothing to a model and depends on no provider's package.

The distribution name is blackboardx; the import name is blackboard. The documentation, including the API reference, is at https://moeinroghani.github.io/blackboardx/.

Install

pip install blackboardx
pip install 'blackboardx[postgres]'     # PostgresStore
pip install 'blackboardx[mongodb]'      # MongoStore
pip install 'blackboardx[notifier]'     # sending notifications to agents over HTTP
pip install 'blackboardx[agent]'        # BoardClient, for an agent calling a blackboard
pip install 'blackboardx[conformance]'  # the suite a store of your own is held to

The base install has no runtime dependency. InMemoryStore holds the record in the process, and SqliteStore uses sqlite3 from the standard library. A deployment keeps the record in the database it already runs, and the adapter for that database needs its driver.

Documentation

Example

from datetime import timedelta

from blackboard import (
    Agent,
    Level,
    Premise,
    RunLimits,
    Settled,
    SqliteStore,
    create_model,
)

notifications = []

model = create_model(
    board_id="incident-4471",
    store=SqliteStore("incidents.sqlite3"),
    regions=[Level("platform"), Premise("window")],
    premises={"window": ["2026-08-16T20:00", "2026-08-16T22:00"]},
    agents=[Agent(name="ocp", notify=notifications.append)],
    limits=RunLimits(wall_clock=timedelta(minutes=10), idle=timedelta(seconds=1)),
)

(notification,) = notifications
window = model.reader.read_premise("window").value
model.control.write("platform", {"window": window, "findings": ["oom"]}, writer="ocp")
model.control.ack(notification.notification_id, agent="ocp")

assert model.control.wait_closed(timeout=timedelta(seconds=10)) == Settled()

Running that example a second time raises DuplicateRegionError, because the board is already in incidents.sqlite3. attach_model opens a run over the board.

License

Apache-2.0. The license text is in LICENSE, and every distribution carries it.