Reasonix

Reasonix is an open-source, MIT-licensed coding agent designed to run locally for long sessions. A single Go engine can be used from the terminal/TUI, desktop app, local browser UI, or a VS Code/VSCodium-compatible editor through ACP. It supports plan-first execution, per-tool permissions, workspace sandboxing, per-turn checkpoints, rewind, project memory, MCP servers, plugins, and skills. Reasonix uses the user's own API keys with configured OpenAI-compatible providers such as DeepSeek or MiMo; the product is not a hosted model subscription, and model/API costs are paid to the selected provider.

✨ Key Features

• Terminal-native agent — Work interactively in the TUI or run headless automation with reasonix run.
• Local engine — Run the coding engine on the user's machine with the configured provider key.
• Four surfaces — Use terminal, desktop, local browser UI, or editor integration through ACP.
• Plan mode — Ask the agent to plan before it touches files or executes tools.
• Tool permissions — Gate reads, writes, and shell commands individually.
• Workspace sandbox — Bound what commands and tools can touch during a session.
• Turn checkpoints and rewind — Review or undo a turn without changing commit history.
• Project memory — Initialize project context and keep append-only session continuity.
• MCP and plugins — Extend the agent with MCP servers, tools, prompts, resources, providers, and sidecars.
• Skills — Add reusable Markdown skills and isolated exploration, research, review, or security-review workflows.
• Editor integration — Connect VS Code, VSCodium, Eclipse Theia, and other ACP-compatible editors.
• Cross-platform distribution — Official builds support macOS, Windows, Linux, and WSL across common architectures.

💳 Pricing

• Open source — Reasonix is MIT-licensed and the public engine/CLI is available as open source.
• Reasonix software — The official site does not present a hosted Reasonix subscription price for the local engine, CLI, desktop app, or browser UI.
• Model usage — Users bring their own DeepSeek, MiMo, or other compatible provider key; provider token charges and model pricing are separate from Reasonix.
• Desktop and editor builds — Official downloads and editor extensions follow the project distribution model; check the release page for current versions.

Do not describe Reasonix as a free hosted AI model or claim that API inference is cost-free. Users remain responsible for provider billing, key security, and any third-party MCP or plugin costs. The local web UI should be protected with token or password authentication before exposing it beyond the local machine.

⚖️ Alternatives Compared

🎯 Use Cases

• Repository refactoring: Plan, edit, test, inspect diffs, and rewind changes in a local repository.
• Bug investigation: Ask the agent to trace failures, inspect logs, and propose a minimal fix.
• Test-driven changes: Run code edits and test commands behind explicit permission gates.
• Long-running tasks: Leave a session running, queue tasks, and resume it later.
• Headless automation: Invoke reasonix run from scripts or pipelines with intentional permissions.
• Editor assistance: Continue an agent session from a VS Code-compatible editor through ACP.
• Private code work: Keep source code on the local machine except for requests sent to the chosen model provider.
• Custom tooling: Add MCP servers, plugins, prompts, and skills for project-specific workflows.
• Team review: Use checkpoints, approvals, and transcript/diff review to make autonomous changes inspectable.

Local execution does not automatically make a workflow safe. Protect provider keys, review every write and shell command, authenticate any exposed browser/serve endpoint, and audit third-party MCP servers, plugins, and skills before installing them.