Features

Khadim is moving from a CLI coding agent into a local-first automation platform. This page separates what works today from the platform surfaces under active development.

Working today

The current CLI agent supports interactive and headless coding workflows.

• Interactive terminal UI with streaming output and live tool execution.
• Batch mode through khadim --prompt.
• Script-friendly execution through khadim exec.
• JSON-line event output for embedding and automation.
• Session save, load, switch, rename, and delete commands.
• Provider and model switching from the UI or command line.
• OAuth login for supported providers such as Copilot and Codex.
• Workspace instruction discovery through AGENTS.md.
• Coding tools for file reads, writes, edits, shell commands, grep, git, web search, delegation, and memory.

Active development

These capabilities exist as previews, scaffolds, or implementation plans.

• Desktop app built with Tauri and React.
• RPA harness for screenshots, screen inspection, and input automation.
• Qwen VLA local experiment for guarded visual UI actions.
• Web app control plane for shared agent runs.
• Docker runtime for isolated local and server execution.
• Memory stores with search, capacity management, and self-improvement nudges.
• WASM plugin system for custom tools and optional desktop UI tabs.

Platform roadmap

The product direction centers on managed, reusable automations:

• Automations are saved runnable tasks promoted from chat or built in an editor.
• Agents are persistent automation personas with instructions, tools, triggers, approval policy, and runner settings.
• Sessions are single executions of an automation or task.
• Connectors bind Khadim to external systems such as email, spreadsheets, browsers, HTTP APIs, and files.
• Environments hold runtime variables and credential bindings.
• Memory lets agents keep durable knowledge across sessions.

Read Architecture to see how these parts fit together.