kyrozen-web serves the browser UI and API through the same runtime, workspace, SQLite store, permission gates, provider configuration, and tool adapters used by the…

docs/api.md READ THE DOCUMENT

Architecture

OpenKyrozen is a modular monolith. Its Python package contains the implementations; root scripts are launchers. The Bubble Tea client remains in the tui Go package.

docs/architecture.md READ THE DOCUMENT

Command and keyboard reference

The Rich command loop, Bubble Tea TUI, and web UI share the agent modes and durable runtime, but their keyboard and session controls differ. Slash commands are entered in…

docs/commands.md READ THE DOCUMENT

OpenKyrozen vs other open-source agents

This is a capability comparison, not a speed, quality, or benchmark claim. It compares the documented product shape of OpenKyrozen, OpenCode, OpenClaw, and Codewhale as…

docs/comparison.md READ THE DOCUMENT

Configuration

OpenKyrozen has separate provider connection settings and workspace behavior settings. Provider credentials come from their provider-specific environment variable, ambient…

docs/configuration.md READ THE DOCUMENT

Development guide

See CONTRIBUTING.md for the issue, contribution, and pull request workflow.

docs/development.md READ THE DOCUMENT

Skills, plugins, and learned artifacts

OpenKyrozen loads reusable guidance from bundled skills, user-installed skill packages, project instruction files, and Python plugins. Their roles and trust requirements…

docs/extensions.md READ THE DOCUMENT

Conversation history and rollback

History preserves conversation snapshots as a tree. Each node is tied to a user actor, workspace, session, and parent version. The current head represents the active line.…

docs/history-and-rollback.md READ THE DOCUMENT

OpenKyrozen documentation

This index covers the current product guides, developer references, and dated engineering records. Start with a task below; use the linked source files when an…

docs/index.md READ THE DOCUMENT

Install, update, and recover

The published v2.0.8 release supports Python 3.12 and 3.13. Python 3.14 is excluded by the package metadata. The official installers provision Python through uv; a source…

docs/installation.md READ THE DOCUMENT

Memory and local storage

OpenKyrozen distinguishes durable facts and event history from retrieval indexes. SQLite is the authoritative store for v2 sessions, events, memory claims, durable tasks,…

docs/memory-and-storage.md READ THE DOCUMENT

Native project intelligence and built-in skills

OpenKyrozen ships Graphify 0.9.64, GitHub CLI 2.101.0 integration, and Ponytail 4.10.0 as versioned built-in skills. /skills reports their installed versions and lifecycle…

docs/native-project-intelligence.md READ THE DOCUMENT

Compact prompts and tool discovery

The prompt profile is optional. The default remains classic.

docs/prompt-discovery.md READ THE DOCUMENT

Model providers

OpenKyrozen separates provider selection, credentials, model defaults, and task-complexity routing. The provider registry in openkyrozen/providers/registry.py is the…

docs/providers.md READ THE DOCUMENT

Engineering reports and evidence

This page indexes records of work performed at a stated revision and time. A report's test count, feature matrix, provider trial, and limitations describe that run. They…

docs/reports.md READ THE DOCUMENT

Security, permissions, and deployment

OpenKyrozen runs local code with the permissions of the operating-system account that launches it. Its capability and approval layers restrict agent requests inside the…

docs/security.md READ THE DOCUMENT

Self-learning, memory, and evidence

Self-learning is one of OpenKyrozen's two defining features. It is a guarded feedback loop around real work: the agent records what happened, proposes a small reusable…

docs/self-evolution.md READ THE DOCUMENT

Sub-agents

OpenKyrozen can delegate independent investigations or specialized work without manual agent creation. The main agent receives delegation guidance at the start and during…

docs/subagents.md READ THE DOCUMENT

Durable tasks and scheduled work

This guide describes resumable work recorded by the local runtime. A normal chat turn and a durable task have different lifecycle rules: accepting a plan or requesting…

docs/tasks-and-automation.md READ THE DOCUMENT

OpenKyrozen runtime inventory

This file is generated from the live runtime registry (AgentRuntime.AVAILABLETOOLS), the FastAPI route table (server.app.routes), and the MCP schema builder. Run make…

docs/tool-inventory.md READ THE DOCUMENT

Troubleshooting

Start by recording kyrozen --version, the operating system, Python version if relevant, the launch command, and the exact sanitized error. Do not paste API keys,…

docs/troubleshooting.md READ THE DOCUMENT

Usage, token accounting, and estimated cost

OpenKyrozen records provider usage locally when the provider returns token counts. The ledger attributes completed calls to the provider and model used, including the…

docs/usage-and-cost.md READ THE DOCUMENT

Usage guide

This is the shortest path from an installed OpenKyrozen command to useful work.

docs/usage.md

Engineering notes and reports

READ THE DOCUMENT

Agentic runtime roadmap

OpenKyrozen now has one durable interaction state machine across the classic CLI, Bubble Tea TUI, web UI, REST, and SSE. Ask and Plan are read-only; accepted plans become…

docs/agentic-runtime-roadmap.md READ THE DOCUMENT

Jev Decision and Decision Assist

Jev Decision is OpenKyrozen's optional judgment layer. It is not the main chat model and it does not execute tools or approve work. Instead, it answers small typed…

docs/decision-assist-validation.md READ THE DOCUMENT

Fast mode benchmark — 2026-09-27 (legacy report)

Fast is now named System One. This historical report is retained for compatibility with earlier raw files and documents the old end-to-end protocol. Use the System One…

docs/fast-mode-benchmark.md READ THE DOCUMENT

Modular refactor validation

Validation date: 2026-09-30. Results cover the modular refactor included with this record. The unrelated, untracked AGENTS.md was preserved.

docs/modular-refactor-validation.md READ THE DOCUMENT

Final production audit — 4 October 2026

This is the historical HOLD snapshot. The shipping follow-up supersedes its unresolved gates with new evidence.

docs/production-audit-2026-10-04.md READ THE DOCUMENT

Production shipping follow-up — 4 October 2026

This report supersedes the earlier HOLD recommendation for the gates rechecked below. The original report and evidence remain a historical record of the unfinished…

docs/production-shipping-audit-2026-10-04.md READ THE DOCUMENT

Self-learning audit — 2026-10-03

This audit traces the 20 switches through their trigger, durable product, and later use. PASS in the matrix means an isolated repeatable scenario exercised that chain,…

docs/self-learning-audit-2026-10-03.md READ THE DOCUMENT

System One benchmark protocol

benchmarks/systemone.py is the repeatable typed-decision benchmark. It uses the same fixed labeled corpus for routing, an explicitly supported clarification, learning…

docs/system-one-benchmark.md

Community and project policies