What is Kition? A 5-minute product tour

The problem Kition is trying to solve

Knowledge work is usually split across incompatible surfaces. Notes live in one application, project trackers in another, source files in a folder, and AI in a separate browser tab. Every transition loses context: you paste text into chat, explain where it came from, copy the answer back, and then manually update the table that tracks the work.

Kition treats the workspace folder as the source of truth for project-owned content. A document can explain a project, a table can track records, a Whiteboard can map the problem, and the Agent can work with selected context. Provider credentials, logs, caches, account sessions, and device preferences stay outside the workspace.

Connected work surfaces

The product is easier to evaluate as several work surfaces that share one project context. The exact feature depth varies by surface, so use a real task and verify the controls you need instead of relying on a checklist copied from an older release.

• Documents: Markdown-based writing and project narrative inside the workspace.
• Tables: .kitable structured records that can support operational data beside the narrative.
• Whiteboards: a visual surface for diagrams, mind maps, flows, and reviewable proposals.
• Browser: an embedded desktop browser for manual navigation, persistent site sessions, and supported page-context extraction.
• Agent: workspace-scoped chats with explicit context, a model picker, streamed tool activity, artifacts, changed-file paths, plans, table Apply steps, and shell approval when the runtime requests it.
• Workflows: visible trigger-and-action automation with testing and run history.

How model choice and credentials work

Choose Kition Cloud or connect OpenAI, Anthropic, DeepSeek, Kimi, or a custom endpoint that implements the supported responses or anthropic_messages wire. Sync the provider model catalog rather than copying model identifiers from an old guide.

In Kition 0.1.41, provider credentials are stored as plaintext files in Kition application data outside the workspace. They are not encrypted or stored in the operating-system keychain. Read Configure an AI provider and Protect provider credentials before entering a production key.

What the agent can and cannot do

The Agent can receive runtime tools for documents, files, tables, web results, embedded-browser context, images, artifacts, and shell commands. Availability changes by provider, model, runtime, pane, and turn. It is suitable for tasks such as summarizing selected documents, drafting structured records, inspecting a browser page, or producing a workspace artifact.

The current safety model is feature-specific. Shell commands can pause for Allow once, a remembered prefix, or Deny. Some table plans require Apply, and changed-file or preview surfaces make supported edits inspectable. Kition 0.1.41 does not have a universal per-tool Permissions pane, public Hooks setup, public MCP setup, or public Subagent workflow. Important changes still require review and independent verification.

A realistic first session

• Install the desktop app from the download page and open a new local workspace.
• Create one Markdown document describing a project, decision, or research topic you already understand.
• Configure a provider using the provider setup guide, then choose Sync now and select a text model.
• Open AI Chat, attach the document, and ask for a read-only summary before requesting changes.
• Create a small table with status, owner, and due-date fields, then ask the agent to add two sample records.
• Inspect the Agent timeline, table plan or changed-file result, and the underlying workspace data before accepting the outcome.

Who Kition is a good fit for

Kition is aimed at developers, researchers, writers, consultants, product teams, and serious personal-knowledge-management users who want one system for writing, structured data, research, and Agent execution.

It may not be the right choice when your primary requirement is large-scale real-time co-editing, public page publishing, fully autonomous browser interaction, or a shipped Hooks/MCP/Subagent control plane. Kition is strongest when a local desktop workspace, mixed document/data surfaces, and reviewable Agent work fit the task.

How to evaluate it with a real task

• Create and reopen a document, a table, and a Whiteboard to confirm the workspace fits your real project structure.
• Run a read-only Agent task with one attached document and inspect every reported tool step.
• Test one small table plan and verify created or updated rows at the source.
• Open a browser page, sign in manually if required, and confirm the Agent can read the intended page context without claiming unsupported clicks or form submission.
• Stop an in-progress task and confirm you understand which completed actions remain.
• Compare the workflow with your current stack using the same real task, not a demo prompt.

Where to go next

For the complete product model, read What is Kition. Then follow Install Kition, Open your first workspace, Configure an AI provider, and Agent chat basics. Feature pages now distinguish shipped behavior from setup contracts that are not exposed in 0.1.41.