Every tool that lets an AI actually do the work rather than just suggest it. Terminal agents, AI IDEs, editor plugins, autonomous software engineers, the frameworks that orchestrate them and the sandboxes they run inside. Vendor neutral, community voted, updated continuously.
Hermes and OpenClaw are two open-source agentic coding projects, both listed below. We borrowed their names for this directory because between them they defined the shape everything else converged on: an agent with a plan, a shell, a set of tools and a way to check its own work, running on weights you choose rather than a service you rent.
That shape now spans official vendor CLIs, community forks, research agents, orchestration libraries and sandbox providers. They compete hard on the surface and quietly converge underneath, mostly around MCP for tools, container isolation for execution and some notion of subagents for parallel work. This directory tracks all of it in one place so you can compare like for like instead of reading twenty landing pages.
You choose the model. No tool in this directory locks you to a single provider without an escape hatch.
The tool plans, acts and checks its own work. Autocomplete-only products live in the coding assistants directory instead.
Open source, an open protocol, or at minimum a documented tool surface you can audit and extend.
AI-first code editor built on VS Code. Understands your codebase, writes code, fixes bugs and now runs background agents in parallel.
GitHub AI assistant. Completion, chat, PR summaries, code review and a coding agent that opens pull requests from issues.
Anthropic official agentic coding CLI. Reads whole codebases, runs tests, commits changes and drives subagents, all from the terminal. Also ships as a desktop app and IDE extension.
Open-source agentic coding tool, and the project this directory takes half its name from. Provider agnostic, runs locally, and built around an agent loop you can read rather than a hosted service you rent.
OpenAI terminal agent for long autonomous coding sessions. Sandboxed by default, with cloud handoff for tasks that outlast your laptop.
Autonomous coding agent for VS Code. Plans and executes complex tasks across files, runs terminal commands and drives a browser. Fully open source.
Open-source agent that grew up alongside OpenClaw and shares much of its tooling surface. Aimed at long-running autonomous work, with message passing between the agent and its tools.
AI-native IDE with the Cascade agent for autonomous multi-file editing and deep codebase understanding. Now part of Cognition.
Anthropic official SDK for building agents on the same harness Claude Code uses. Subagents, hooks, skills, MCP and permission control.
Open-source AI code assistant for VS Code and JetBrains. Bring any model, local or hosted, and build custom assistants from config.
AI pair programming in your terminal. Git-native, edits multiple files with clean commits and works with essentially any model you can point it at.
Graph-based orchestration for stateful, long-running agents. Durable execution, checkpointing and human-in-the-loop interrupts.
Google open-source terminal agent with a generous free tier and a very large context window. Strong at reading big repositories in one pass.
Open-source autonomous software engineer. Runs in a container, browses, writes code and iterates against tests until the task passes.
Browser-based full-stack AI development environment. Prompt to running app to deploy without installing anything.
OpenAI lightweight framework for multi-agent workflows with handoffs, guardrails and built-in tracing.
Community fork of Cline with custom agent modes, per-mode model routing and a large prompt-customisation surface.
Vercel AI UI generator. Natural language to React components, then copy the code or deploy it straight away.
High-performance collaborative editor written in Rust, with a built-in agent panel and first-class support for external agent protocols.
Role-based multi-agent framework. Define crews with goals and backstories, wire them to tools and let them delegate between each other.
Agent built into the Replit workspace. Plans, writes and deploys an app inside a hosted environment with no local setup.
Open-source Cursor alternative. Run local models or cloud APIs with no telemetry. Full VS Code fork with complete privacy.
Prompt-to-product builder that wires a database, auth and deployment behind the generated front end.
Microsoft framework for conversational multi-agent systems, with an event-driven core and a visual builder for prototyping.
Cognition hosted autonomous engineer with its own workspace, browser and shell. Aimed at delegated tickets rather than pair programming.
Open-source terminal coding agent with a provider-agnostic core, a TUI, and a server mode you can drive from anything.
Cloud sandboxes purpose-built for AI agents. Sub-second Firecracker microVM starts, filesystem and network isolation, SDKs in Python and TypeScript.
Agentic terminal that turns the shell itself into the surface for AI work, with multi-agent panes and shared session context.
Serverless compute with a sandbox primitive that agents can call directly. Useful when the agent needs GPUs, not just a shell.
Princeton research agent that turns any language model into a software engineer through a purpose-built agent-computer interface. The reference implementation for SWE-bench work.
Type-safe agent framework from the Pydantic team. Structured output validation is the core primitive rather than an afterthought.
Open-source developer agent from Block. Extensible through MCP servers, runs fully locally and handles long multi-step tasks.
Self-hosted AI coding assistant. Runs entirely on your own hardware with no external calls, which makes it the default for air-gapped teams.
Open-source VS Code agent that merges ideas from Cline and Roo, with built-in provider routing and no markup on model costs.
Very fast completion with a 1M token context window. Optimised purely for low-latency inline suggestions.
JetBrains own coding agent, wired into IntelliJ, PyCharm, GoLand and the rest of the suite with full IDE index access.
TypeScript-first agent framework with workflows, evals, RAG primitives and deployment adapters for the usual edge platforms.
Context-engine-first assistant aimed at very large codebases. Indexes the whole repo and keeps that index warm across sessions.
Sourcegraph agentic coding tool built around unconstrained token use and deep code search across very large monorepos.
Fast, stateful sandbox infrastructure for agent workloads, with snapshot and resume so a long task can survive a restart.
AWS coding assistant with deep AWS service knowledge, transformation jobs for legacy upgrades and a usable free tier.
Adaptive AI IDE with a builder mode that scaffolds whole projects from a spec, plus a generous free model allowance.
Test and review focused assistant. Generates meaningful test suites and does structured PR review rather than free-form chat.
Planning-first coding engine for large changes. Keeps work in a protected draft sandbox so you review before anything touches your branch.
The open spec most agent harnesses fall back on for reproducible local environments. Not glamorous, but it is the shared substrate.
Open-source autonomous agent with a self-hosted server option, strong on refactoring and repository-wide edits.
AI-native IDE forked from VS Code with proactive suggestions, inline edits and automatic debugging loops.
Try adjusting your search or filter.
Start with a terminal agent. Claude Code and Codex CLI are the most capable, Aider is the most model-agnostic, opencode and Goose are the ones you can read end to end.
Editor plugins keep your existing setup. Cline and its forks give you the most control, Continue gives you the widest model choice, Copilot gives you the deepest GitHub integration.
Look at autonomous SWE agents. OpenHands if you want to self-host and inspect everything, Devin if you would rather buy the outcome than run the infrastructure.
Pick an orchestration framework by language and by how much state you need. LangGraph for durable long-running graphs, the Claude Agent SDK for the Claude Code harness, Mastra or Pydantic AI for typed stacks.
Do not skip the sandbox layer. E2B and Daytona give you microVM isolation with sub-second starts, Modal adds GPUs, dev containers cover the local case.
Tabby self-hosts completely, Void and Continue run against local models, and most terminal agents will happily point at an Ollama or vLLM endpoint.