Trending Tools

Claw Code is a Rust CLI harness for running Claude-style terminal agents with explicit auth, workspace builds, and parity checks for teams that want local control instead of a hosted UI.

Claw Code delivers a Rust-based agent harness runtime with plugin extensibility and REPL interface, surpassing 121k GitHub stars within days of launch.

MemPalace keeps full local conversation history searchable instead of asking an LLM to guess what matters, which preserves debugging context and long-term recall.

MemPalace stores raw AI conversations locally, then makes them searchable with a palace-style hierarchy and semantic retrieval instead of summary-first memory loss.

Clean-room Python reimplementation of Claude Code's agent harness architecture, ported from exposed TypeScript snapshot using OmX orchestration, achieving functional parity with 31.7k GitHub stars as of April 2026.

Python porting workspace of Anthropic's exposed Claude Code agent with CLI for status summaries and unittest verification.

Open Design turns local coding-agent CLIs into a deterministic, local-first design pipeline that emits sandboxed artifacts instead of disposable mockups.

OpenClaude lets you run one terminal-first coding agent across cloud and local model backends, with provider profiles, MCP, slash commands, streaming output, and headless gRPC integration.

graphify turns code, docs, PDFs, and screenshots into a provenance-tagged knowledge graph so assistants answer architecture questions from structure, not grep output.

Claude Code Best V5 recreates Claude Code’s terminal workflow in an open-source Bun/TypeScript CLI, so teams can swap Anthropic for compatible providers without losing agentic coding flows.

Career-Ops builds an agentic pipeline on Claude Code to score job offers across 10 weighted dimensions, generate ATS-optimized CV PDFs, scan portals like Greenhouse and Lever, and track applications with data integrity checks.

colleague.skill turns scattered chats, docs, and exports into a reusable AI skill that preserves a teammate’s working style after they leave.

Claurst is a Rust-native terminal coding agent that gives you Claude Code-style workflows with multi-provider support, local control, and no telemetry.

caveman strips agent replies to high-signal fragments and cuts about 75% of output tokens without losing the technical fix.

OpenMythos recreates a looped RDT-style transformer in PyTorch, letting you test recurrent-depth reasoning, MLA/GQA attention, and sparse MoE routing without building the architecture from scratch.

EmDash replaces WordPress-era plugin risk with schema-first content, sandboxed Worker isolates, and a TypeScript-native Astro integration.

NVIDIA NemoClaw deploys sandboxed OpenClaw agents in a k3s-orchestrated environment with OpenShell runtime, routing inference to NVIDIA cloud APIs in under 5 minutes setup.

free-code is a telemetry-stripped, open-source Claude Code fork that removes Anthropic guardrails, unlocks experimental flags, and adds multi-provider support in a Bun-based terminal CLI.

AutoResearchClaw generates conference-ready academic papers from a single prompt via 23 autonomous stages, integrating real literature from OpenAlex, Semantic Scholar, and arXiv.

OpenHarness gives you an inspectable open-source agent harness with tool use, memory, permissions, and multi-agent orchestration in one Python codebase.

Lark CLI turns Lark/Feishu workspace automation into 200+ commands and 19 AI Agent Skills, so developers can script Messenger, Docs, Base, Sheets, Calendar, Mail, Tasks, and Meetings without custom API glue.

Claw Code Parity delivers an autonomously built coding harness with 6.5k GitHub stars that ports Project Claw Code to Python while preparing Rust integration via AI-driven parallel sessions.

女娲.skill turns public material into explicit persona models and decision heuristics, so Claude Code can reason with documented boundaries instead of vague roleplay.

TurboQuant+ isolates KV-cache compression as a measurable engineering problem, showing that you can cut cache footprint by 3.8-6.4x without paying the quality penalty that usually comes from key compression.

Open Multi-Agent turns a single goal into a dependency-aware agent DAG, runs independent work in parallel, and returns typed results without forcing you to hand-wire orchestration.

Bun-powered TUI fix of leaked Claude Code source enables local execution with custom Anthropic APIs, multi-agent orchestration, MCP plugins, and Python-bridged desktop control.

A thin CDP-based browser harness that lets LLM agents self-heal by writing missing helpers mid-task instead of failing on edge cases.

Delivers a 200-page source code analysis of production AI agents plus a minimal Python CLI agent with fake LLM for offline structure testing.

GBrain turns an AI agent into a stateful system by storing life context in a local-first markdown knowledge base backed by PGLite and pgvector, then reading and writing that brain on every interaction.

Claude Code Haha fixes leaked Claude Code source for full local Ink TUI execution with any Anthropic-compatible API endpoint.

LingBot-Map turns long image streams into interactive 3D reconstructions with feed-forward inference and paged KV-cache attention, avoiding the slow optimization loops that make classic SfM pipelines unusable for live mapping.

boneyard extracts pixel-accurate skeleton states from real UI so teams stop hand-crafting placeholders and ship loading screens that match production layouts.

A curated GPT-Image-2 prompt library that replaces blind trial-and-error with reusable, dated examples, localized documentation, and source-backed visual references.

fireworks-tech-graph turns plain-language architecture specs into editable SVG and print-ready PNG diagrams, removing manual diagramming while preserving UML, AI agent flows, and semantic layout discipline.

ClawTeam spawns swarms of AI agents that self-organize to execute complex tasks like ML experimentation and full-stack development from a single CLI command.

Converts any website, Electron app, or local tool into a CLI via Chrome session reuse and AI-powered command discovery.

ex-skill parses WeChat logs, QQ exports, and photos to generate Claude-compatible AI skills that replicate ex-partner speech patterns and shared memories with 5-layer persona structures.

Cursor Cookbook turns Cursor's SDK into copyable TypeScript patterns for shipping agents, scripts, and workflows without reverse-engineering the API.

CodeBurn reads local sessions from Claude Code, Cursor, Codex, OpenCode, Pi, and GitHub Copilot, then turns them into per-project token, cost, and one-shot success analytics without a wrapper, proxy, or API key.

OpenSpace equips AI agents with self-evolving skills that auto-fix breaks, capture successful patterns, and share improvements across agents, cutting token usage by 46% on real tasks.

CodeBurn turns local AI coding session files into a provider-agnostic token, cost, and retry dashboard so you can see exactly where assistant spend is being burned.

OpenMAIC transforms any topic or document into interactive multi-agent classrooms with AI teachers, quizzes, simulations, and TTS whiteboard in minutes via Next.js backend.

Recordly delivers native macOS ScreenCaptureKit and Windows DXGI screen capture with automated cursor smoothing and zoom animations in an Electron app.

Open Agent SDK TypeScript executes full agent loops in-process without subprocesses or CLIs, supporting Anthropic and OpenAI-compatible APIs across cloud, serverless, Docker, and CI/CD deployments.

Understand Anything uses Claude's multi-agent pipeline and static analysis to generate an interactive knowledge graph mapping every file, function, class, and dependency in a 200k+ line codebase.

OmniVoice turns short reference audio into zero-shot speech across 600+ languages at RTF 0.025, so you can ship multilingual voice cloning without training a custom TTS model.

Open CoDesign turns prompts into editable artifacts locally with BYOK model support, on-device version history, and exportable HTML, PDF, PPTX, ZIP, and Markdown outputs.

Bumblebee turns messy on-disk developer metadata into read-only NDJSON inventory and exact exposure matches without executing package managers or touching source files.

Mirage collapses dozens of remote services into one bash-native filesystem so agents can copy, grep, snapshot, and compose work across backends without juggling SDKs.

wterm gives browser apps a DOM-backed terminal with near-native WebAssembly parsing, native selection and copy-paste, and WebSocket PTY transport without the usual accessibility trade-offs.