Skip to content

Junjo Python SDK 0.69.0

Choose a tag to compare

View all tags
@mdrideout mdrideout released this 20 Sep 20:38
sdk-python-v0.69.0
330c74b
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Junjo is a Python library for building AI Agents and explicit Workflows with observable application state. This release lets them work with the same live Store, so a tool's state changes can be followed through the whole execution in Junjo AI Studio.

Breaking upgrade: start with fresh Studio data

Use this SDK with Junjo AI Studio 0.85.0. Telemetry contract 3 replaces contract 2; older semantic consumers are unsupported.

Stop the old stack and wipe Studio application data, or select a new empty data directory. Users, API keys, developer tokens, evaluations, and trace history are not migrated. Complete first-user setup and configure new credentials after the upgrade. docker compose down --volumes does not clear host-mounted data: follow the tagged reset procedure.

Python 3.11 requires 3.11.9 or newer; Python 3.12 requires 3.12.3 or newer. Python 3.13 and newer remain supported. Explicit Agent/Tool/context/result generic declarations must use the current application State/Store type parameters.

What you can do

  • Give an Agent or Workflow its own application Store, or pass a shared live Store through execute(store=...). Tools access it through their execution context; results retain detached state snapshots.
  • Trace which Node or Tool changed a field, inspect the state available at a decision, and follow shared state across nested Agent/Workflow executions. Agent runtime bookkeeping stays private; Subflows still use isolated Stores with explicit pre/post mapping.
  • Start from the new OpenAI SDK example: an order-support Agent, a direct Node tool, a conditional Workflow tool, typed state, and OpenInference model telemetry through one OpenTelemetry provider.
  • Find supported setups through the examples/integrations index and expanded composition, model-driver, and provider-instrumentation documentation.

The example Studio URL is corrected, and AI Chat now generates a name before passing that identity into biography generation. Intentional evaluation failures remain part of the demonstrations.

Validation includes the full SDK suite on supported minimum Python patches and newer interpreters, package validation, example tests, live Gemini/Grok checks, and Studio state/trace/CLI evidence parity. The OpenAI integration was tested with the real client and instrumentor using deterministic HTTP responses; a paid OpenAI provider call was not part of release validation.

Changelog · Changes since SDK 0.68.0 · Matching Studio release

Assets 2
Loading