DeepSeek-Reasonix
by
esengine

Description: DeepSeek-native AI coding agent for your terminal. Engineered around prefix-cache stability — leave it running.

View on GitHub ↗

Summary Information

Updated 34 minutes ago
Added to GitGenius on June 21st, 2026
Created on April 21st, 2026
Open Issues & Pull Requests: 1,063 (+0)
Number of forks: 1,642
Total Stargazers: 26,355 (+1)
Total Subscribers: 74 (+0)

Issue Activity (beta)

Open issues: 881
New in 7 days: 219
Closed in 7 days: 113
Avg open age: 6 days
Stale 30+ days: 142
Stale 90+ days: 0

Recent activity

Opened in 7 days: 174
Closed in 7 days: 99
Comments in 7 days: 424
Events in 7 days: 1,371

Top labels

  • v2 (1,430)
  • bug (1,349)
  • enhancement (752)
  • windows (703)
  • desktop (577)
  • agent (574)
  • v1 (374)
  • macos (203)

Repository Insights (GitGenius)

Median issue/PR response: 0.0 hours
Mean response time: 6.3 hours
90th percentile: 1.6 hours
Tracked items: 2,578

Most active contributors

Detailed Description

DeepSeek-Reasonix is a terminal-based AI coding agent written in Go that integrates with DeepSeek's language models and other OpenAI-compatible endpoints. The project is engineered specifically around DeepSeek's prefix caching mechanism to maintain low token costs during extended sessions, allowing users to leave the agent running continuously without prohibitive expense.

The repository underwent a significant architectural shift with version 1.0, transitioning from a TypeScript implementation to a ground-up Go rewrite. The main development branch is now main-v2, while the earlier 0.x TypeScript releases remain on the v1 branch in maintenance-only mode. Despite this major rewrite, the npm installation command remains consistent, delivering the Go binary for version 1.0.0 and above while legacy TypeScript builds are available for older versions.

The core design philosophy centers on configuration-driven architecture and composability. All major components—providers, the agent itself, enabled tools, and plugins—are declared in a reasonix.toml configuration file rather than hardcoded into the application. This approach enables multi-model support where DeepSeek ships as a preset but any OpenAI-compatible endpoint can be added as a configuration entry. The system supports optional two-model setups where an executor and planner run in separate, cache-stable sessions for enhanced reasoning capabilities.

Plugin architecture follows the MCP-compatible standard, with external tools running as subprocesses over stdio JSON-RPC. Built-in tools self-register at compile time, and the system maintains cache-aware context by injecting a small stable environment summary at startup. Stale tool output is intelligently pruned before summary compaction, with the built-in tool schema contract documented for regression review.

Distribution is designed for zero friction. The application compiles as a single static Go binary with CGO_ENABLED=0, eliminating external dependencies beyond a TOML parser. Cross-compilation to six targets (darwin and linux on amd64 and arm64, plus Windows variants) occurs with a single command. Prebuilt archives and SHA256SUMS are available on every GitHub release, with Windows builds code-signed through SignPath Foundation's free certificate program.

GitGenius activity data reveals intense development velocity. The repository has processed 2498 issues and pull requests with a median response latency of 0.0 hours and mean of 6.5 hours, indicating rapid community engagement. The most active labels are v2 with 1352 items, bug with 1287, and enhancement with 739. Primary contributors esengine, Bernardxu123, and SivanCola have driven 2109, 1003, and 886 events respectively. The project ranks in the top tier across multiple categories—top 2 in Agents by velocity, top 3 in LLMs by velocity, and top 3 in CLI by velocity according to oosmetrics.

The project maintains comprehensive documentation including configuration guides, migration instructions from the legacy TypeScript version, tool contracts, checkpoint and rewind functionality for edit safety, and bot integration guides for platforms like Feishu, Lark, and WeChat. A bilingual Discord community provides setup assistance and feature discussion.

DeepSeek-Reasonix
by
esengineesengine/DeepSeek-Reasonix

Repository Details

Fetching additional details & charts...