Skip to main content

Kenaz Harness

The Kenaz Harness is the desktop AI application you install and run on your own machine. Open it, pick an AI provider, and start chatting — ask it to read files, run searches, connect to services via MCP, and more. Everything it does is written to a local audit log so you always have a record of what happened.

The Harness is open source: github.com/kameas-ai/kenaz-harness. It runs standalone, or inside the Kenaz Workbench if you prefer an all-in-one sandboxed environment.

:::tip Ready to get started? Jump straight to the Quickstart — you'll be chatting in about five minutes. :::

What you can do with it

  • Chat with any AI provider you choose. Bring your own API key for Anthropic, OpenAI, OpenRouter, AWS Bedrock, Azure OpenAI, Google Gemini, Ollama, or any OpenAI-compatible endpoint. Your keys stay in your OS keychain — they never reach a Kameas server. See Providers.
  • Give the model real tools. Built-in tools cover filesystem reads/writes, shell commands, and web search. Connect any MCP server to add GitHub, Slack, Postgres, Notion, and hundreds more. Every tool call is gated by a permission prompt before it runs — you decide what the model is allowed to do.
  • Organize work into sessions and projects. Conversations are first-class objects: branch them, attach files, group them into projects, search across all of them. See Sessions.
  • Build up shared context over time. Long conversations compact automatically; useful facts pin into long-term memory; team-wide guidance goes into signed context packs that land in every teammate's session.
  • Keep a tamper-evident record of everything. Every prompt, response, tool call, and permission decision lands in a hash-chained local audit log before it's shown in the UI.
  • Stay local-first — connect to Fleet only when you want to. By default the Harness is fully self-contained and nothing is sent to Kameas. Optionally sign in to Kenaz Fleet to receive pushed team configuration, a shared tool catalog, and settings sync across devices. See Connecting to Fleet.

How it's organized

The window has a left rail of primary views and a Settings hub for provider, security, and authoring configuration:

Left rail

ViewWhat it's for
SessionsChat with the model. New conversations, branches, projects.
ToolsConnected MCP servers and the built-in tool registry.
WorkflowsAgentic multi-step recipes you can run or customize.
ContextsContext packs — guidance injected into every matching session.
MemoryInspect and edit what the model remembers across turns.
ArtifactsFiles the model produced during a session.
Audit logBrowse and export the local audit log.
SettingsAll configuration — providers, permissions, hooks, secrets, and more.

Settings hub (reached via Settings in the left rail)

SectionWhat's inside
ProvidersAdd / remove AI providers and pick the default model.
BundlesInstalled bundles — signed packs of providers, skills, and MCP recipes.
SecretsManaged secret store for tool credentials and environment values.
PermissionsWhat's currently allowed and per-session permission grants.
PolicyCedar-based policy rules pushed from Fleet or authored locally.
HooksPre-send / post-send shell or built-in hooks that run on every turn.
CompactionTune how long conversations are summarized.
Slash CommandsCustom slash-command library.

Where to next

New here? Follow the onboarding path:

  1. Quickstart — install, add a provider, send your first message (5 minutes).
  2. Using the Harness — sessions, tools, memory, and daily patterns.
  3. Features — per-feature reference for everything the Harness can do.
  4. Providers — detailed setup for each AI provider.
  5. Connecting to Fleet — optional team features.

Something not working? See Troubleshooting & FAQ.