# gitclaw

> A universal git-native multimodal always learning AI Agent (TinyHuman)

- **URL**: https://www.freshcrate.ai/projects/gitclaw
- **Author**: shreyaskapale
- **Category**: AI Agents
- **Latest version**: `v1.4.3` (2026-04-22)
- **License**: MIT
- **Source**: https://github.com/open-gitagent/gitclaw
- **Language**: TypeScript
- **GitHub**: 211 stars, 36 forks
- **Registry**: npm (`gitclaw`)
- **Tags**: `agent`, `ai`, `cli`, `git`, `llm`, `npm`, `sdk`

## Description

A universal git-native multimodal always learning AI Agent (TinyHuman)

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v1.4.3` | 2026-04-22 | High | ## What's New  Fixes the **\"Your session hit the maximum duration of 60 minutes\"** failure in voice mode — previously, once OpenAI terminated the Realtime session, every subsequent message returned \"server had an error while processing your request\" and the agent was stuck until restart.  ### Proactive refresh On \`session.created\`, a 55-minute timer is armed. When it fires, the adapter tears down the WS and opens a fresh one with the same instructions, tools, and voice. Users notice a sub- |
| `1.4.1` | 2026-04-21 | Low | Imported from npm (1.4.1) |
| `v1.4.2` | 2026-04-20 | High | ## What's New  ### New \`edit\` built-in tool The agent can now modify files without rewriting them in full. Works on both the local filesystem and inside the E2B sandbox.  **Modes:** - **Exact string replacement** (default) — must match uniquely; errors with a clear message if not found or ambiguous - **replace_all** — replace every occurrence - **regex** — treat \`old_string\` as a JavaScript regular expression; \`new_string\` can use \`\$1\`-style backreferences, and custom flags (\`i\`, \`m\ |
| `v1.4.1` | 2026-04-17 | High | ## What's New  Enhances the Logs UI with much deeper server-side instrumentation so that even the smallest error becomes visible.  ### Global safety net - `uncaughtException`, `unhandledRejection`, and Node `warning` handlers — captures crashes that would otherwise disappear silently - Error objects now logged with full stack traces  ### HTTP layer - Every `/api/*` request logged with method, path, status, and duration - All 4xx/5xx responses logged at the corresponding level - `jsonReply` auto- |
| `v1.4.0` | 2026-04-17 | High | ## What's New  ### Unified Logs Tab A new **Logs** tab in the web UI that surfaces all server-side logs in real time — voice, Telegram, WhatsApp, triggers, SDK, scheduler, and system events.  **Features:** - Real-time log streaming via WebSocket - Filter by source (voice, telegram, whatsapp, triggers, sdk, scheduler, system) - Filter by level (info, warn, error) - Full-text search - Auto-scroll with toggle - Color-coded source badges and level-based message coloring - REST endpoint (`GET /api/lo |
| `v1.1.0` | 2026-03-15 | Low | ## What's New  - **Cron Scheduler** — schedule prompts to run on a cron expression or one-time datetime, with JSONL logging, repeat/once modes, and auto-disable for one-shot jobs - **Scheduler UI** — new Scheduler tab in the voice dashboard with form, preset cards, toggle/run/delete controls - **6 REST API endpoints** — `/api/schedules/list`, `save`, `delete`, `toggle`, `run`, `logs` - **Installer Auto-Resume** — returning users with an existing `~/assistant` setup skip the interactive wizard an |
| `v1.0.0` | 2026-03-15 | Low | ## SkillsFlow — v1.0.0 (Major Release)  ### What's New - **SkillFlows Builder** — Visual workflow builder with a skills panel, step cards, and arrow connectors - **Drag-and-Reorder** — Drag step/gate cards to rearrange flow order with a blue drop indicator - **Approval Gates** — Pause flows and request approval via Telegram or WhatsApp before continuing - **Speaker Mute** — New button to mute/unmute agent audio output instantly - **New Workflow Button** — Quick-start a fresh workflow from the Sa |
| `v0.4.1` | 2026-03-14 | Low | ## What's Changed  - Removed all scaffolding artifacts (flappy bird, calculator, hello, jumping ball, architecture diagrams, requirements.txt) - New `install.sh` with pixel sprite banner and two setup modes:   - **Quick Setup** — OpenAI + Anthropic + optional Composio, launches in 30 seconds   - **Advanced Setup** — voice adapter, model, project dir, Composio, Telegram, port selection - Auto-opens browser on launch (macOS, Linux, Windows) - Saves API keys to `.env` for future runs  ### Full inst |
| `v0.3.0` | 2026-03-04 | Low | ## Local Repo Mode with Session Branches  Adds `--repo <url> --pat <token>` to auto-clone a GitHub repo locally, create a session branch, run the agent, auto-commit changes, and push the session branch.  ### Features - **CLI**: `gitclaw --repo <url> --pat <token> "prompt"` — clones, branches, runs, commits, pushes - **Session resume**: `--session gitclaw/session-<id>` checks out existing branch with full context - **SDK**: `query({ repo: { url, token } })` — same flow programmatically - **Token |
| `v0.2.0` | 2026-03-04 | Low | ## What's new  **Sandbox mode** — Run agent tools inside an isolated E2B cloud VM via [gitmachine](https://github.com/open-gitagent/gitmachine). The agent (LLM calls) still runs locally; only tool execution is remote. All changes are automatically committed to a session branch.  ### Usage  **CLI:** ```bash export E2B_API_KEY="..." export GITHUB_TOKEN="ghp_..." gitclaw --dir ~/my-project --sandbox ```  **SDK:** ```typescript import { query } from "gitclaw";  for await (const msg of query({   prom |

## Citation

- HTML: https://www.freshcrate.ai/projects/gitclaw
- Markdown: https://www.freshcrate.ai/projects/gitclaw.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/gitclaw/deps

_Generated by freshcrate.ai. Indexes npm releases for AI-agent ecosystem packages._