aclade

The official Python SDK for Aclade — autonomous AI experts that do real work.

import aclade and you can list your AI team, delegate tasks to individual experts, and read their reports — from any Python code, notebook, or pipeline.

Docs & examples: github.com/aclade-ai-expert/aclade-docs — SDK reference, architecture notes, and worked examples.

Install

pip install aclade

Python 3.8+, zero third-party dependencies (standard library only).

Get a token

1. Sign in at aclade.com
2. Settings → Local Agent → Get connection token (the same token the local connector uses)
3. export ACLADE_TOKEN=<TOKEN>

Tokens are signed 12-hour sessions bound to your workspace. If you already connected a machine with the npm connector, the SDK picks the token up from ~/.aclade-agent.json automatically — import aclade just works.

Quickstart

import aclade

team = aclade.Aclade()                    # token from $ACLADE_TOKEN or ~/.aclade-agent.json
team = aclade.Aclade(token="<TOKEN>")     # …or explicitly

# Who's on the roster?
for e in team.experts():
    print(e["name"], "—", e["title"])

# Delegate a task and wait for the report
result = team.delegate("Priya", "Summarize our top 3 customer complaints and draft replies")
print(result["status"])                   # completed | needs-input | needs-approval
print(result["run"]["report"])

The three outcomes

Note on execution. delegate() sends the task to the platform's task runner, which plans and — when the task can be served from what the expert already knows — rehearses the procedure, but does NOT drive your local machine. A report that begins with the ⚠️ Planning brain — no live execution banner is telling you exactly that. For real execution (filesystem, shell, browser, Jira/SAP/CRM connectors), run the same task in the expert's chat in the portal with the local connector attached.

CLI

aclade experts                # roster
aclade run "Priya" "task"     # delegate + print the report
aclade tasks                  # pending tasks by expert
aclade agent                  # local-agent fleet (connector installs)
aclade whoami                 # identity carried by the token

Where does the AI run?

On aclade.com — not on your machine. This SDK is the official client: it holds your token, sends work, and returns results. The reasoning harness, tools, and per-second metering are Aclade platform services — the same way import stripe never contains Stripe's rails.

Optionally, your machine can also run the zero-trust local connector (npm i -g aclade), which lets your experts act on your own filesystem, shell, and browser — outbound HTTPS polling only, no inbound ports.

Security notes

• The token is a credential: anyone holding it can delegate work as your workspace (billed to your account). Keep it in an env var or secrets manager.
• This package ships no AI logic, model weights, or telemetry.
• License: proprietary (UNLICENSED), for users of the Aclade platform.