Recently updated open-source projects

GitHub Copilot CLI brings its coding agent into a local command-line workflow for building, editing, debugging, and refactoring code. It can also work with repositories, issues, and pull requests through an authenticated GitHub account and extend its tools through MCP servers. Actions are previewed for approval, and use requires an active Copilot subscription with prompts counting against premium-request quotas.

PicoClaw packages a personal AI assistant as a portable Go binary for x86, ARM, RISC-V, MIPS, and other architectures, with a reported core memory footprint below 10 MB. It can connect to many model providers and chat channels, manage MCP server configurations, schedule jobs, and spawn asynchronous sub-agents. Most hosted model choices still require their corresponding credentials.

Start with a rough software idea, refine it through a design conversation, and move into implementation, or point the tool at an existing project for repeated improvement cycles. It tracks experiments locally, evaluates each change before keeping or reverting it, and can run through Claude Code, Bob Shell, or OpenAI Codex.

Galley runs Claude Code, Codex, or GLM tasks in managed Git worktrees, records commands and diffs, and asks a supervisor to accept, retry, escalate, or stop each attempt. Repository profiles define checks, scope, secrets policy, and optional pull-request handoff, while evidence remains on disk for later inspection. Task YAML is trusted execution input, so Galley should be used only with task authors and repositories you trust.

Configure each price, version, rule, or deadline with its owning source page and the local files that cite it. fact-drift saves a baseline, extracts the current value, and shows the old value, new value, source quote, and stale matches before changing anything. Updates require approval, while login-only pages, heavy JavaScript sites, and PDFs need a manual or separate browser-based check.

yt-playlist-subs creates one ordered Markdown transcript from a YouTube playlist. It prefers creator subtitles, skips machine-translated tracks, uses local faster-whisper for videos without captions, and accepts an exported member cookie for restricted videos; each entry carries a source badge and silent clips receive an extra transcription safeguard.

Emdash is a desktop environment for running coding agents in parallel, with each task isolated in its own Git worktree for reviewing diffs, CI checks, pull requests, and merges. It supports local projects and SSH/SFTP remote machines, and keeps app state in local SQLite. Emdash does not upload code or chats to its servers, but the selected agent CLIs may still send content to their own model providers.

Every Code is a community fork of OpenAI Codex for local terminal development. It extends the upstream tool with browser actions and multi-agent workflows, while Auto Drive and background Auto Review support longer execution and non-blocking checks. It supports ChatGPT sign-in or API-key use, can orchestrate external CLIs such as Claude and Gemini, and is not an official OpenAI product.

h5i hooks into Claude Code or Codex to version human prompts, agent context steps, and trimmed tool output as recoverable Git objects. It can also place several agents in isolated worktrees, coordinate peer review, test competing implementations, and apply the selected result. The workspace evidence lives inside repository refs, so it can be pushed, fetched, inspected offline, and used without depending on a hosted development environment.

webapp-last-90 detects a web app’s stack, scores an 11-dimension readiness baseline, and gates launch on ten binary blockers such as real auth, tenant isolation, rate limits, legal pages, production monitoring, CI/CD, health checks, and a rehearsed backup restore. It separates autonomous work from items needing account access or a decision, then returns SHIP-SAFE or NOT-YET with evidence.

D2 Diagrams for Claude Code turns an architecture request into D2 code, renders an SVG, and checks the visual result before delivery. Its render loop looks for tangled edges, drifting labels, imbalance, and clipped text, using ELK by default and TALA when a valid license is available. The plugin works without TALA; PNG self-inspection may download a headless browser on first use.

Forge brings an AI coding workflow to the terminal through an interactive TUI, one-shot commands, or a Zsh colon-prefix interface. Its built-in forge, sage, and muse agents separate implementation, read-only code research, and planning. Developers can retain conversation context, choose among many model providers, and connect MCP tools without leaving their command-line workflow.

San is a native Go terminal harness distributed as a single binary of roughly 12 MB. It lets you switch models and search backends, attach plugins or MCP servers, and control actions through ask, auto-accept, or autopilot modes. Sessions can resume or fork, and a local inspector can replay transcripts; provider credentials are still required for the services you choose.

Continue provides a coding agent through a CLI, a VS Code extension, and a JetBrains plugin, with documented configuration and customization options. The repository has reached its final 2.0.0 release and is now read-only and no longer actively maintained. Existing users can still run it, but the project recommends the CLI over the JetBrains plugin.

claude-mem captures tool-use observations, creates semantic summaries, and makes relevant project context available in later sessions. It adds layered retrieval, history search, a web viewer, and private tags for excluding sensitive content; installation must register its hooks and worker through the provided installer or plugin flow, not only install the SDK.

Open Interpreter is a Rust terminal coding agent designed for lower-cost and open models such as DeepSeek, Kimi, Qwen, and GLM. Its TUI can switch models and agent harnesses, execute commands inside native sandboxes, and use a built-in QA skill to operate and test web or native applications. Configuration and session state remain on the local machine.

LiteLLM works as either a Python SDK or a self-hosted AI gateway, exposing more than 100 providers through a unified OpenAI-compatible interface. Teams can switch among services such as OpenAI, Anthropic, Gemini, Bedrock, and Azure without rewriting every integration, while centralizing virtual keys, spend tracking, load balancing, logging, and guardrails at the gateway.

n8n combines a visual workflow builder with JavaScript, Python, and npm packages for automations that connect models, data, and business systems. It can run in n8n Cloud or be self-hosted with Node.js or Docker, with a large library of integrations and templates. The source is available under a fair-code licensing model rather than a conventional permissive open-source license.