Skip to content

CHANGELOG.md

Latest commit

 

History

History
85 lines (68 loc) · 4.05 KB

CHANGELOG.md

File metadata and controls

85 lines (68 loc) · 4.05 KB

Changelog

All notable changes to RustClaw are documented in this file.

[0.5.0] - 2026-04-12

Added

  • Cross-platform installer — one-line install for macOS, Linux, and Windows
    • curl -sSL .../install.sh | sh (macOS/Linux)
    • irm .../install.ps1 | iex (Windows PowerShell)
  • CI/CD release pipeline — GitHub Actions builds 5 targets on tag push (aarch64-apple-darwin, x86_64-apple-darwin, x86_64-linux, aarch64-linux, x86_64-windows)
  • BFCL benchmark for qwen3-coder:30b — 98.9% across 1,000 questions at 2.6s/q (4.3× faster than qwen2.5:32b)
  • Gemma4 writing test — 6-topic Chinese writing comparison across 3 models; gemma4:26b confirmed as best Chinese writing model
  • Vendored OpenSSL feature — optional vendored-openssl Cargo feature for Linux ARM cross-compilation

Changed

  • Default recommended model updated to qwen3-coder:30b (MoE, 18 GB) from qwen2.5:32b (dense, 19 GB)
  • README updated with dual-model BFCL benchmark table, one-line install instructions
  • Translated READMEs (zh-TW, zh-CN, ko) synced with all v0.5.0 changes
  • Roadmap updated: removed Web UI and WASM plugins, added cross-platform install, marked multi-model routing as in-progress

Fixed

  • Resolved all 15 compiler warnings (zero-warning build)
  • Removed dead code: MAX_HISTORY_MESSAGES, load_workspace_context(), ProcessInfo struct, unused imports

Removed

  • Stale benchmark files (v4 timeout-fix, v5-pro, v6 failed run, redundant simple_python subset, outdated BENCHMARK_REPORT.md)
  • Web UI and WASM plugin system from roadmap (Web UI covered by existing 5 interfaces; WASM would break binary size budget)

[0.4.0] - 2026-04-10

Added

  • BFCL 99.7% — 1,000 questions on official Berkeley Function Calling Leaderboard (qwen2.5:32b)
    • simple_python: 99.75% (399/400)
    • multiple: 99.5% (199/200)
    • parallel: 100% (200/200)
    • parallel_multiple: 100% (200/200)
  • T-Eval 95.5% — 2,146 questions across 4 core categories (retrieve 98%, plan 96%, review 96%, instruct 92%)
  • Polished README with badges, comparison table, Key Advantages section
  • Marketing docs for Reddit and awesome-list submissions

Changed

  • README redesigned with benchmark tables and architecture diagram
  • Internal benchmark v3→v5 improvement: 81% → 97%

[0.3.0] - 2026-04-08

Added

  • R-Mem integration — replaced internal memory system with rustmem crate
    • Three-tier memory: vector recall + graph (entity-relation) + audit history
    • Mixed-mode scoping: local + user + global
    • Integer ID mapping (prevents LLM UUID hallucination)
  • 6 Claude Code-inspired optimizations — understand-first tool ordering, history compression, workspace context, error retry hints
  • 500-question internal benchmark with per-category scoring

Changed

  • src/session/ reduced from ~600 lines to 219 lines (thin wrapper over rustmem)
  • Memory features delegated upstream to R-Mem crate

[0.2.0] - 2026-04-07

Added

  • MCP client — stdio transport, tool auto-discovery and transparent routing
  • System monitoring — process list, docker status, PM2 status, HTTP ping, system stats
  • Email — IMAP inbox reading + SMTP sending (Gmail compatible)
  • Cron scheduling — system check, email scan, GitHub scan on configurable schedules
  • Enhanced GitHub tools — repo scanning + auto-PR from issues
  • Persistent sessions — SQLite-backed conversation history

Changed

  • Tool count expanded from 8 to 22

[0.1.0] - 2026-04-06

Added

  • Initial release
  • Gateway WebSocket server (OpenClaw-compatible protocol)
  • Agent runner with Anthropic + OpenAI (Ollama) provider support
  • SSE streaming with tool calling agentic loop (max 10 iterations)
  • Telegram channel (long polling, streaming edit, ACL)
  • Discord channel (@mention, server management commands)
  • 8 built-in tools: read_file, write_file, patch_file, list_dir, search_code, run_command, process_list, process_check
  • TOML config with environment variable overlay
  • 14 dangerous command patterns blocked in run_command
  • CLI with clap subcommands: gateway, agent, status, health, github