activity
Recent activity
The repository ships fast. This is a static snapshot from the README news log — see the changelog and the live feed on GitHub for the current state.
- release v1.6.0 — ClawCodex Web in a browser tab, and --nano, a deliberately smaller harness #839–#886
- feat ClawCodex Web — a three-column browser client on the same gateway the desktop app already used #839–#872
- feat Trajectory — every model request and tool call as a metered ledger, with per-step tokens and timings #839, #849, #869
- feat clawcodex --nano — six tools and a ≈2K-token fixed payload instead of ≈17K #879–#886
- stat Nano A/B on terminal-bench 2.1: 2/2 solved in both modes — $0.0042 nano vs $0.0186 default (deepseek-v4-flash, k=1)
- release v1.5.0 — ClawCodex Desktop, a native app on the same agent core as the terminal #802–#808
- feat ClawCodex Desktop — streaming chat, permission approvals, and a sidebar that resumes TUI sessions #802–#805
- feat clawcodex serve — one loopback port serving /api/* REST plus a JSON-RPC WebSocket gateway #803, #804
- feat macOS preview build — DMG and zip packaged with electron-builder #805
- release v1.4.0 — fusion models give a text-only model vision, GPT-5.6, 30 providers #766–#787
- stat Codebase stats: 1,269 Python files, 302,697 lines of Python
- feat Fusion models — a text-only base model plus a borrowed vision model #771
- feat groq, cerebras, baseten and xai — the provider registry reaches 30 #784
- feat /mode becomes /permissions with a three-level picker, Full Access by default #768
- release v1.3.0 — 80.9% on Terminal-Bench 2.1 (Opus 5), persistent memory, VS Code extension #720–#764
- stat Terminal-Bench 2.1: 72/89 solved — 80.9% pass@1 (k=1) on claude-opus-5, would slot ≈3rd on the public board
- stat Codebase stats: 1,250 Python files, 289,355 lines of Python
- feat Claude Opus 5 support + interactive /effort fix on Anthropic #746
- feat Bounded persistent memory with a background self-improvement review #731
- feat VS Code extension driving the agent-server over stdio #727
- release v1.1.0 — run OpenAI & Claude models on your subscription, not metered API billing #697–#702
- stat Codebase stats: 1,200 Python files, 271,599 lines of Python
- feat Sign in with ChatGPT — OpenAI models via a ChatGPT Plus/Pro subscription #698
- feat Claude Pro/Max subscription login over OAuth (PKCE), with the moved-endpoint repair #697
- feat Meta (api.meta.ai) provider + 1M-context muse-spark-1.1 reasoning model #692
- feat /plan mode with implicit plan-mode entry/exit, ported from the reference #676
- feat --worktree/-w — parallel sessions isolated in separate git worktrees #672
- release v1.0.0 — goal-directed autonomy, hooks & MCP wired for production, hardened permissions #580–#668
- stat Codebase stats: 1,170 Python files, 256,909 lines of Python
- feat /goal + /subgoal completion-condition loop — the agent works until an LLM judge confirms the goal #664
- feat Monitor tool — stream long-running shell output with backpressure #665
- feat MCP OAuth auth flow — OAuth-protected MCP servers are now reachable #662
- feat 'clawcodex mcp serve' — re-expose ClawCodex tools as an MCP stdio server #635
- release v0.7.0 — TUI auto-theming, faithful inline rendering, Claude-Code-style tool trail #573–#577
- release v0.6.0 — interactive TUI REPL parity with the ink reference #383
- feat One-click installer — curl -fsSL https://clawcodex.app/install.sh | bash
- feat 18 new LLM providers — the registry grows 7 → 25 #377
- feat DeepSeek becomes the default provider; prefix-cache headline added across the READMEs #365–#367
- feat DeepSeek prefix-cache exploitation — byte-stable prefix makes cache-hit input over 200× cheaper #363
- feat Z.ai GLM-5.2 support — new zai provider for the GLM Coding Plan #343
- stat Codebase stats: 1,093 Python files, 233,520 lines of Python
- feat Dynamic workflow engine + /deep-research wired end-to-end #262–#271
- feat Web tooling refresh — Tavily-backed WebSearch, rebuilt WebFetch #265
- feat Interactive command-system parity (/theme, /model, /mcp, /tasks, …) #230–#261
- feat Remote-bridge parity + CLI transports (phases 0–18) #200–#226
- feat /advisor token-efficient coding mode — cheap worker + expensive reviewer #181–#193