Act in your workspace
Read and change project files, run commands, work with Git, browse the web, and inspect project context through guarded tools.
Explore capabilitiesOPEN SOURCE · LOCAL-FIRST · OUTCOME-LED
An agent that learns from verified work.
OpenKyrozen works in your project, remembers useful context, and improves repeatable workflows from evidence. You choose the model. Its capabilities stay bounded by explicit safety controls.
OPENKYROZEN / TUIPROJECT CONTEXT
BUILT FOR REAL WORK
From a request to a result you can inspect.
Use one terminal-native agent for coding, research, and project work. OpenKyrozen keeps useful state locally and records what actually happened, so progress can build on more than a good-sounding answer.
THREE PARTS, ONE WORKFLOW
Act. Remember. Improve.
Keep useful context
Sessions, tasks, memory claims, and learning evidence use durable local storage, so a new conversation need not start from zero.
See how it remembersLearn from outcomes
It can turn repeated, verified workflows into reusable skills. Learning cannot silently expand permissions or approve actions.
Read the learning modelA PRACTICAL LEARNING LOOP
Experience becomes evidence before it becomes a habit.
Do the work
OpenKyrozen uses the tools allowed for the active mode and workspace.
Record what happened
Outcomes, corrections, tool receipts, and acceptance evidence provide a traceable basis for learning.
Reuse only what holds up
Candidate improvements pass evidence gates; they do not grant new capabilities or rewrite model weights.
CONTROL IS PART OF THE DESIGN
Helpful autonomy, with clear boundaries.
Ask and Plan stay read-only. Agent actions pass through capability and approval checks. Learning can improve how work is approached, but it cannot grant permissions or approve its own work.
Read the safety and learning detailsSTART WITH YOUR OWN WORKSPACE
Bring an agent into the work you already do.
Install OpenKyrozen, choose a provider, and start with a project task.