Overview
The Kenaz Harness is the desktop application — the part of Kenaz you actually launch and click. It runs on your machine, talks to whichever AI provider you configure, manages sessions and tools, and writes everything to a hash-chained audit log.
If you just want to start using it, jump to the Quickstart.
What it does
- Brokers AI inference. You bring your own API key for one or more providers (Anthropic, OpenAI, OpenRouter, AWS Bedrock, Ollama, or any OpenAI-compatible endpoint). Keys live in your OS keychain — never on a server we control. See Providers.
- Runs tools. Built-in tools (filesystem, shell, web search) plus any Model Context Protocol (MCP) server you point it at. Every tool call is gated by a permissions prompt before it runs.
- Manages sessions. Conversations are first-class objects with branches, projects, attachments, and per-session artifacts. Switch between them, fork them, share them. See Sessions.
- Manages context across the elevation ladder. Five rungs from working memory in a single chat all the way up to signed context packs shared across your team or organization. Long conversations compact automatically; durable facts pin into long-term memory; team-wide guidance lands in every teammate's session via signed packs. Move information up the ladder when it earns its keep.
- Writes an audit log. Every prompt, response, and tool invocation lands in a tamper-evident local log before it leaves the harness. See Audit log.
- Stays on your machine. No telemetry. No prompts shipped to us. Outbound network is locked down at the webview layer; the only outbound traffic is to the AI provider you configured.
How it's organized
The window is laid out as a left rail of views and a main canvas:
| View | What it's for |
|---|---|
| Sessions | Chat with the model. New conversations, branches, projects. |
| Providers | Add / remove AI providers and pick the default model. |
| Tools | Connected MCP servers and the built-in tool registry. |
| Permissions | What's currently allowed and per-session permission grants. |
| Hooks | Pre-send / post-send shell or built-in hooks that run on every turn. |
| Memory | Inspect and edit what the model remembers across turns. |
| Corpora | RAG document collections you can attach to a session. |
| Artifacts | Files the model produced during a session. |
| Settings | Dials for the harness — keyboard shortcuts, autonomy, logging, update channel. |
| Audit | Browse and export the local audit log. |
Where to next
- First time using it? Read the Quickstart — it walks you through install, picking a provider, and your first message in about five minutes.
- Stuck on a provider setup? Each provider has its own page under Providers.
- Want to know what's actually shipping? Features is the per-feature index.