powered by DataGOLFrom goal to outcome /
it takes a city.
Open source.
MIT licensed.
Explore models, tools, memory, permissions, and more using the concept buttons below.
Visit DataGOLWhy HarnessX
Fluent to build.
Flexible by design.
Compose agents in Python, plug in the models and tools you need, and extend their behavior. Memory, coordination, control, and visibility come together in one SDK.
Fluent constructs
Typed functions. Clear configuration.
Python you can read and compose.
Pluggable components
Choose your model. Bring your tools.
Connect services through MCP.
Room to extend
Add skills, hooks, and middleware.
Package features as extensions.
Your agent.
Your rules.
Choose what runs, what needs approval, and where work stops. Set per-tool permissions and budgets before execution.
Pick up where you left off.
Keep conversations, notes, and long-term facts. Save and reload a session when the work continues.
Give every task a specialist.
Give each specialist a role and tools. Delegated tasks get fresh memory and return results to the lead agent.
Understand every step.
Follow lifecycle events and inspect recorded runs. Export a verified incident for offline playback.
Continue after a restart.
Persist run state, tool results, and approval waits. Resume unfinished work in the same session after your application restarts.
Typed judgments for your workflow.
Pass shared state and independent Choice, Score, or Noul questions to Jev. It returns typed answers in one batch; your code decides what follows.
Build once.
Extend any agent.
Package tools, middleware, hooks, and context as one reusable feature. Install it in any HarnessX agent.
Keep your app in the loop.
Observe run, model, and tool events. Connect callbacks to logs, metrics, and progress updates.