Reasonix
A terminal coding agent tuned for DeepSeek prefix-cache stability in long-running development sessions.
Per-dimension scores and reasoning
Evidence shows documented design for permission checks, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects control, rollback, and source attribution. However, no implementation details are provided, and publisher is unverified, so deductions apply.
Documentation and CI config indicate self-consistency, dependency availability, and failure message handling, but no actual run evidence, so deductions apply.
Documentation describes multiple use scenarios, capability boundaries, trigger precision, and environment fit, but no concrete configuration examples, so deductions apply.
Information architecture is clear, install notes are detailed, naming is stable, examples and FAQ exist, known limitations are partial, license is clear, versioning has changelog, maintenance responsibility is clear, but no full changelog, so deductions apply.
Output usability, marginal value, and cost-benefit are documented, but no actual performance data, so deductions apply.
Claims are traceable, cross-source corroboration exists, facts and inferences are separated, but no specific test results, so deductions apply.
- Publisher is unverified; treat source with caution.
- Static review cannot verify actual runtime behavior; dynamic testing is needed.
- Dependency security is only implied by CI config; no vulnerability scan results provided.
What does this agent do, and when should you use it?
Reasonix is a terminal-based coding agent with an interactive CLI/TUI and non-interactive `reasonix run` execution. Its `reasonix.toml` configuration declares providers, models, the agent, enabled tools, and plugins; DeepSeek is a preset, while any OpenAI-compatible endpoint can be configured. The tool system combines built-in tools registered at compile time with external subprocess plugins communicating over stdio JSON-RPC and described as MCP-compatible. Its context approach uses a small stable startup environment summary and removes stale tool output before summary compaction to preserve cache stability. Distribution includes a static Go binary cross-compiled for six platform and architecture targets, a desktop application, and a VS Code extension backed by a local CLI process.
Run reasonix setup to configure a provider and model, then start an interactive session with reasonix or submit a one-off task with reasonix run "implement the TODOs in main.go". Providers, the agent, enabled tools, and plugins are declared in reasonix.toml; an executor model and planner model can optionally operate in separate cache-stable sessions. Reasonix invokes built-in tools compiled into the program and can run external tools as subprocesses over stdio JSON-RPC. Within an interactive session, /init creates project instructions. The VS Code extension starts the local reasonix acp backend and adds chat, editor context, tool-call approvals, model selection, and workspace sessions.
- A solo developer who wants to hand a terminal agent a bounded task such as implementing TODOs in
main.gocan usereasonix run. - A developer running lengthy DeepSeek-assisted coding sessions who wants context maintenance designed around prefix-cache stability.
- A team using an internal or self-hosted OpenAI-compatible endpoint that wants to configure it rather than add provider-specific code.
- A user who wants separate executor and planner models for a coding workflow while keeping their sessions distinct.
- A VS Code, VSCodium, or Eclipse Theia user who has installed the local CLI and wants editor chat, workspace sessions, and tool-call approvals.
What are this agent's strengths and limitations?
- Centralizes providers, models, tools, and plugins in
reasonix.toml, with DeepSeek as a preset and any OpenAI-compatible endpoint configurable without new code. - Its documented context maintenance specifically targets prefix-cache stability through a stable environment summary and stale tool-output pruning before compaction.
- Supports both compile-time registered built-in tools and external MCP-compatible subprocess tools over stdio JSON-RPC.
- Offers a static Go binary for darwin, linux, and windows on amd64 and arm64, alongside CLI/TUI, desktop, and editor delivery paths.
- You must configure a provider and model before use; the supplied documentation does not specify the credential format or authentication procedure.
- The VS Code extension does not bundle the CLI, so it requires a locally installed Reasonix engine and launches
reasonix acpitself. - External plugins run as subprocesses over stdio JSON-RPC, creating an operational need to manage plugin executables and their environments.
- The README documents OpenAI-compatible endpoints but provides no evidence of dedicated adapters or feature parity for other named model vendors.
How do you install or deploy this agent?
For the CLI/TUI, run npm i -g reasonix, or on macOS run brew install esengine/reasonix/reasonix. Next, run reasonix setup to configure a provider and model; the README does not document the exact credential fields or authentication flow. To build from source, run git clone https://github.com/esengine/DeepSeek-Reasonix.git, cd DeepSeek-Reasonix, then make build, which produces bin/reasonix(.exe). For editor use, install the CLI first, then install the VS Code extension with ID SivanLiu.reasonix-agent.
How do you use this agent?
After reasonix setup, run reasonix for an interactive session or reasonix run "implement the TODOs in main.go" for a one-off task. Use /init inside an interactive session when you want Reasonix to create project instructions. For advanced setup, declare providers, the agent, enabled tools, and plugins in reasonix.toml; external plugins run as stdio JSON-RPC subprocesses.
FAQ
Is Reasonix limited to DeepSeek?
Why does it emphasize long-running sessions?
Can it use external tools?
Can the VS Code extension run without the CLI?
reasonix acp backend.