Clawcodex

ClawCodex is an open source coding agent for the terminal, written as a Python native rebuild of the Claude Code architecture that works with many model providers instead of Claude alone. It keeps a genuine tool calling loop, a streaming REPL, persistent sessions and markdown based skills, and puts one runtime in front of 30 providers, including Anthropic, OpenAI, Z.ai GLM, MiniMax, OpenRouter and DeepSeek. It suits developers who want Claude Code style agentic coding while choosing their own vendor, region and price tier. More than 30 built in tools cover files, bash, web access, subagents and MCP. The same agent runs in the terminal, in a desktop app and in a browser tab served from one local gateway, so a session started in the terminal can be resumed in the browser, where a Trajectory view breaks down tokens, time to first token, throughput and cache use for every step. Skills are defined in SKILL.md files with named arguments and per skill tool limits, and images can be mentioned inline for vision capable models. Version 1.7.0 rebuilt the multi agent layer with a session supervisor, persistent named agent teams sharing a dependency aware task board, and background workers that resume with their history. The project emphasizes token efficiency; it states that prefix cache reuse makes long DeepSeek sessions more than 200 times cheaper than running a Claude flagship model. ClawCodex is free and MIT licensed, and a one line script installs it on macOS, Linux and WSL. Users pay their chosen model provider directly for usage. The project ships weekly and reports an 80.9 percent score on a public terminal coding benchmark when paired with a Claude Opus model.