Skip to main content

bma-executor-sdk (Python)

PyPI

Connect your machine (hands) or your own agent (brain) to BMA — the executor, harness and agent ends of the bma.executor protocol. bma-sdk drives and uses agents (REST / façade client); bma-executor-sdk connects hands or brains to BMA.

asyncio executor (ExecutorClient), harness (Harness) and agent ends of the BMA Executor Protocol bma.executor.v1.1, the counterpart of the Go executor module. Protocol types are TypedDicts generated into bma_executor_sdk/generated.py from the contract in contracts/bma/executor, so messages are plain JSON dicts.

Version 1.0.0 (tag v1.0.0), see CHANGELOG
Protocol bma.executor.v1.1 (contract 1.1.0, lock.json)
Python 3.10 or newer; depends on httpx and websockets
Integration guide docs/executor (中文)
pip install bma-executor-sdk==1.0.2

The REST / façade client is the separate package bma-sdk.

Executor quick start

import asyncio, os
from bma_executor_sdk import BashInput, BashResult, ExecutorClient, ToolContext

async def main() -> None:
    client = ExecutorClient(
        base_url=os.environ["BMA_BASE_URL"],  # required: see below
        environment_key=os.environ["BMA_ENVIRONMENT_KEY"],
        environment_id=os.environ["BMA_ENVIRONMENT_ID"],
        executor={"name": "my-executor", "os": "linux", "arch": "arm64", "version": "0.1.0"},
    )

    @client.tools.bash.handler  # one typed slot per built-in tool; only registered tools are advertised
    async def bash(input: BashInput, ctx: ToolContext) -> BashResult:
        proc = await asyncio.create_subprocess_shell(
            input["command"], stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE)
        out, err = await proc.communicate()
        return BashResult(stdout=out.decode(), stderr=err.decode(), exit_code=proc.returncode or 0)

    await client.start()
    await client.wait_closed()  # raises when the client ends for good

asyncio.run(main())

Set base_url to your BMA deployment: the hosted service is in early access and the default DEFAULT_BASE_URL (https://api.bma.beeos.ai) does not resolve yet. It is the deployment origin, the same value as BMA_BASE_URL in bma-sdk (for example https://api.staging.bma.beeos.ai); the client posts the environment key to {base_url}/v1/executors/token and drops a trailing /api/v1. The SDK does not read BMA_BASE_URL itself. A token_source carries its own channel URL.

A complete version that runs bash and stops the process on cancellation is examples/executor_minimal.py; set BMA_HARNESS_URL (ws://… or unix:///path.sock) to connect it to a local harness instead of BMA.

Behavior matches the Go SDK:

  • A handler returns a string, content blocks, ToolResult(content, is_error, artifacts) or (bash) BashResult, rendered like the Go Core; raise ToolError(text) reports a tool-level error (is_error: true), any other exception is outcome: "failed". The harness calls tools as await session.tools.bash.call({"command": "ls"}).
  • tool.cancel cancels the handler's task and reports outcome: "canceled"; duplicate tool.call ids run once.
  • tool.result retries -32009 / -32004 / retryable errors and hands a lost result to on_tool_result_lost.
  • Reconnect resumes with resume.executor_id / active_sessions, until stop() by default (backoff 1 s doubling to 60 s, jitter); only close codes 4003 and 4004 stop the client. A positive max_reconnect_attempts bounds the attempts. await client.wait_closed() returns after stop() / executor.close and raises ChannelClosed (4003/4004), CredentialRejectedError (the token endpoint answered 401/403 on reconnect: revoked or invalid credential, not retried) or ReconnectExhaustedError; let it end the process so a supervisor can restart it.
  • Call session_ready after setup and session_ended after cleanup. client.on_request(methods.PROCESS_START, handler) serves process.*, terminal.*, skill.install or workspace.backup yourself with the contract's TypedDicts; await client.call(methods.SESSION_EVENT, params) / await client.notify(methods.TOOL_PROGRESS, params) send typed methods (bma_executor_sdk.methods). There is no execution Core in Python.
  • Input missing a required member fails the call before the handler runs; a tool.call for a tool without a handler is refused with -32012 before the ack; shell / files are enabled while a handler of the group is registered unless set enabled: False, and a handler registered after start() sends executor.capabilities.update. BashResult beyond INLINE_RESULT_BYTES (or truncated=True) is cut to Go Core's preview. The computer_* tools get slots once the contract defines their input.

Harness quick start

from bma_executor_sdk import Harness

harness = Harness(authenticate=authenticate)  # authenticate(ConnectRequest) -> Principal; omit only in-process or on a private socket
server = await harness.serve("127.0.0.1", 0)   # or await harness.serve_unix(path)

executor = await harness.wait_executor()
session = await executor.assign_session({"session_id": "ses_1"})
await session.ready()
result = await session.call_tool({"tool": "bash", "input": {"command": "ls"}})
mcp = await session.mcp("echo")  # raw MCP JSON-RPC
await session.stop("done")

Semantics match the Go Harness: call_tool is idempotent by tool_call_id; sessions and pending calls survive a reconnect within the lease; lease expiry fails pending calls with ToolOutcomeUnknownError.

Transports

Chosen by the channel URL of the token source:

Transport Executor Harness
WebSocket (ws://, wss://) default await harness.serve(host, port)
WebSocket over a Unix domain socket (unix:///path.sock) token_source=static_token("unix:///path.sock", token) await harness.serve_unix(path)
Windows named pipe (npipe:////./pipe/<name>, Windows ProactorEventLoop) token_source=static_token("npipe:////./pipe/<name>", token) await harness.serve_npipe(name)
In process dialer=harness.inprocess_dialer() —

The executor-listens local socket is not implemented in Python (use the Go or TypeScript SDK); named pipes need the Windows ProactorEventLoop (the asyncio default there) and are tested on a Windows runner in CI. See local connection.

Agent executors

client.serve_agent(handler, capability) serves the agent capability; handler implements open_turn, run_turn and close_session (async; cancellation is task cancellation), and run_turn uses turn.emit, await turn.delta(...), turn.inputs (an asyncio.Queue) and turn.request_interaction. The harness side is Harness(agent=platform) with an AgentPlatform. See agent capability.

Go ↔ Python

Go (bmaexecutor) Python (bma_executor_sdk)
NewClient(Options{...}) ExecutorClient(environment_id=..., executor=..., ...)
Options.TokenSource, StaticToken(url, token) token_source=, static_token(url, token)
Client.Start / Stop / State / ExecutorID / ConnectionID / ActiveSessions start() / stop(reason) / state / executor_id / connection_id / active_sessions
Tools.Bash.Handle(fn) @client.tools.bash.handler
OnSessionAssign / OnSessionStop / OnDrain on_session_assign / on_session_stop / on_drain
Requests.X.Handle(client, fn) / Calls.X.Call(ctx, client, p) / Notifications.X.Notify(ctx, client, p) on_request(methods.X, fn) / await call(methods.X, p) / await notify(methods.X, p)
ReportToolResult / SessionReady / SessionEnded / OpenMCPChannel report_tool_result / session_ready / session_ended / open_mcp_channel
ServeAgent(handler, AgentCapability) serve_agent(handler, capability)
AgentHandler.OpenTurn / RunTurn / CloseSession open_turn / run_turn / close_session
AgentTurn.Emit / EmitWithID / NewEventID / Delta / Flush / Inputs() / RequestInteraction / CancelInteraction emit / emit_with_id / new_event_id / await delta / await flush / inputs / request_interaction / cancel_interaction
NewHarness(HarnessOptions{...}), http.Handler Harness(...), await serve(host, port) / serve_unix(path) / inprocess_dialer()
HarnessOptions.Authenticate(*http.Request) (Principal, error) authenticate(ConnectRequest) -> Principal (raise to refuse)
Harness.Executors / WaitExecutor executors() / wait_executor()
Executor.AssignSession / Drain / Capabilities / Protocol assign_session / drain / capabilities / protocol
Executor.StartTurn / SendTurnInput / CancelTurn / CloseAgentSession / RespondInteraction / AgentCapability start_turn / send_turn_input / cancel_turn / close_agent_session / respond_interaction / agent_capability()
Session.Ready / CallTool / CancelTool / Stop / MCP / Done / EndReason ready() / call_tool / cancel_tool / stop / mcp / wait_ended(), ended / end_reason
AgentPlatform AgentPlatform (Protocol: acked_sequence, agent_event, agent_delta, interaction_requested, interaction_canceled)
ErrNotConnected / ErrClientStopped / ErrSessionEnded / ErrToolOutcomeUnknown / ErrAssignRejected / *RPCError NotConnectedError / ClientStoppedError / SessionEndedError / ToolOutcomeUnknownError / AssignRejectedError / RPCError
Core, NewProcessCapability, ListenUnixSocket, Harness.DialExecutor not provided

Examples

Run from examples after uv sync (or pip install bma-executor-sdk):

Example What it shows
executor_minimal.py executor answering bash calls, against BMA or a local harness
harness_loopback.py brain and hand in one process, in-process or --unix PATH
agent_minimal.py agent capability with an in-memory platform

Tests

uv sync && uv run pytest -q

tests/test_executor_interop.py runs the Python executor against the Go harness (WebSocket and Unix socket) and the Python harness against the Go executor; it needs Go.

Metadata

Release files for bma-executor-sdk 1.0.2

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for bma-executor-sdk 1.0.2
File Size Uploaded Provenance Actions
bma_executor_sdk-1.0.2.tar.gz 67.7 kB With provenance Details for bma_executor_sdk-1.0.2.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for bma-executor-sdk 1.0.2
File Interpreter ABI Platform Actions
bma_executor_sdk-1.0.2-py3-none-any.whl Python 3 none any Details for bma_executor_sdk-1.0.2-py3-none-any.whl

Total release size: 124.1 kB

Release files / bma_executor_sdk-1.0.2.tar.gz

Download URL bma_executor_sdk-1.0.2.tar.gz
Size 67.7 kB
Tags Source
SHA-256 checksum
How to use checksums
3ac083d31eb0c4d08226484d9fd4c9bb2f183ab8ce4f8d4a072d4a92d5f98d8c
BLAKE2b-256 checksum
How to use checksums
030897c051700cd4874d809ac4d029152839cca996f447c4e094a9317810ca2b
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 7, 2026.

Transparency log

Release files / bma_executor_sdk-1.0.2-py3-none-any.whl

Download URL bma_executor_sdk-1.0.2-py3-none-any.whl
Size 56.4 kB
Tags Python 3
SHA-256 checksum
How to use checksums
0cee8d2fb09a73ea1b388f69bdce8744d48d94ee30178440ea4e07c2d10e02d0
BLAKE2b-256 checksum
How to use checksums
6fa840de0e6fc9a69c6d431e0701702c11d8398c57e7368fd81365b72812b57d
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 7, 2026.

Transparency log

Security for bma-executor-sdk 1.0.2

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for bma_executor_sdk-1.0.2.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 7, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page