Junjo docs

Junjo gives your coding agent the building blocks and shared evidence needed for recursive self improvement: investigate a failure, test a targeted change, compare the outcomes, and show you what happened.

Two pieces fit into your existing stack:

• Junjo Python SDK lives in your application. It provides instrumentation, workflow and specialist-agent building blocks, local evaluation tooling, and a CLI and skills for your coding agent.
• Junjo AI Studio runs as lightweight containerized services. It stores telemetry, evaluation datasets, results, and execution history for investigation by coding agents and humans.

Your coding agent makes and tests changes. Your application runs the models, tools, and evaluators. Studio connects the outcomes to the execution evidence. Keep your supported agent framework, or build fully with Junjo.

Examples and integrationsChoose a runnable setup for your Agent runtime, model SDK, telemetry, and application state.

Run your first recursive self improvement cycleStart with a failure, describe the improvement to your coding agent, and inspect the evidence behind the result.

Get started

Start with one application feature you want to understand and improve.

1. Set up Junjo AI Studio.

   Deploy Studio to store traces, datasets, and evaluation results in an environment you control.

2. Connect your application.

   Configure OpenTelemetry traces. The Python SDK provides native Workflows and Agents with additional execution and state diagnostics. See the integration guides in the sidebar for supported external runtimes.

3. Connect your coding agent.

   Install the evaluation skill matching your installed SDK. Your agent uses Junjo’s CLI to discover the application’s targets, manage datasets, run experiments, and inspect evidence.

4. Improve one real feature.

   Follow the first improvement guide. Declare a target and its evaluation criteria, establish a baseline, and compare a code or prompt change against the same locked dataset. Open the linked Studio evidence to review the result together.

Explore the platform

Explore alternative architecturesTest different ways to solve the same problem, compare their execution evidence, and choose what to develop next.

Datasets and evaluation runsCreate repeatable scenarios, run application targets locally, and compare outcomes and execution evidence.

Investigate with Junjo AI StudioLet your coding agent and your team inspect the same outcomes, traces, and state changes.

Python building blocksImprove focused capabilities with workflows, specialist agents, concurrency, and observable state.

Keep your agent frameworkObserve OpenAI Agents SDK applications and expose Junjo capabilities as tools. Python integration.

Python API referenceGenerated reference for the public Python API.

Python is the currently released SDK. The first-party framework integration is for the OpenAI Agents SDK for Python. Other framework adapters are planned; receiving standard OpenTelemetry traces does not imply a native adapter for every framework or language.