Open Source
Explore the latest AI open-source projects from GitHub and HuggingFace.
Explore the latest AI open-source projects from GitHub and HuggingFace.
DeepSeek TUI is a terminal-based coding agent for DeepSeek V4 models, written in Rust, that exploded onto GitHub Trending in May 2026 by adding more than 21,000 stars in a single week. The project, by Hmbown, sits at v0.8.30 with a quick release cadence and represents one of the most interesting open-source coding agents released this year: a serious alternative to Claude Code and OpenAI's Codex CLI, but built around DeepSeek's models and shipped as a polished TUI rather than a wrapper around a hosted product. ## A Real Coding Agent in the Terminal DeepSeek TUI is not a chat box that happens to live in a terminal. It is a full coding agent with the same loop modern hosted agents have: read the workspace, plan, edit files, run commands, evaluate, and iterate. What makes it distinct is the level of polish applied to the terminal experience using the ratatui framework. Streaming reasoning blocks render live, tool calls show their arguments and outputs inline, and approval gates pause the agent before any destructive operation. ## Auto Mode for Model and Thinking Level DeepSeek V4 exposes multiple reasoning intensity levels. Picking the right one per task matters for both cost and latency: a quick lint fix does not need the deepest thinking budget, but a refactor across three files does. DeepSeek TUI's auto mode selects both the model variant and the thinking level for each interaction, making the right tradeoff automatically. Users can override per session if they want fixed behavior. ## 1M-Token Context with Prefix Cache Awareness The agent supports DeepSeek V4's 1M-token context window, but more importantly it is prefix-cache aware. When working on a single project across many turns, large portions of the context (file contents, system prompt, workspace layout) repeat. Prefix cache awareness ensures those repeated prefixes hit cache rather than being recomputed, which is a real cost win for long sessions and a feature most coding agents still do not implement correctly. ## Plan, Agent, and YOLO Modes Three operating modes give users control over autonomy. **Plan** mode produces a structured plan and waits for approval before acting. **Agent** mode is the standard loop with approval gates on file edits and shell commands. **YOLO** mode disables approval gates for sandboxed workflows where speed matters more than safety. The mode is a top-level toggle, which is cleaner than burying autonomy settings inside config files. ## Full Tool Suite and MCP The agent ships with file operations, shell execution, git, web search, and MCP server support. The MCP integration matters because it lets users plug in the tool ecosystem that has grown around Claude Code, Cursor, and other MCP-aware agents, without DeepSeek TUI needing to reinvent each integration. LSP diagnostics are surfaced inline, so the agent sees real type errors and warnings from the same language servers the user's editor uses. ## Session Save and Workspace Rollback Sessions can be saved and resumed across restarts, with a durable task queue that survives crashes. More importantly, workspace rollback gives the user a one-command undo on the agent's edits, which is essential for trust during longer sessions where the user is not watching every action. ## Skills, Cost Tracking, and Headless API A skill system enables composable instruction packs, similar to the patterns popularized by Claude Code's recent skills work. Cost tracking shows per-request token spend with a cache hit/miss breakdown, so users can actually see whether their prefix caching is working. An HTTP/SSE runtime API exposes the agent for headless use in CI or longer-running automated workflows, and prebuilt Docker images make deployment straightforward. ## Where It Fits DeepSeek TUI is the best open-source coding agent for users who want to run against DeepSeek models, and one of the most polished terminal-native coding agents available regardless of model choice. MIT licensed, Rust-fast, with a v0.8.x cadence that says this project is moving quickly. If you have been waiting for a serious open alternative to closed coding agents, this is one to install today.
OpenClaw is an open-source, local-first AI gateway with 366K GitHub stars that routes AI responses through WhatsApp, Telegram, Slack, Discord, iMessage, Teams, and 15+ other platforms — zero cloud dependency.
OpenClaw
Open-source personal AI assistant connecting to 13+ messaging platforms with local gateway architecture, voice support, and multi-agent routing.