Find the open-source solution that fits.

Project overview

OpenMontage lets an AI coding assistant orchestrate complete video productions through readable pipeline manifests, stage skills, and Python tools. It handles research, proposals, scripts, scene planning, generated or retrieved assets, narration, music, captions, editing, composition, and post-render checks across explainers, documentaries, demos, podcast clips, localization, and other formats. It can analyze a reference video, use free local or open-footage paths, or connect cloud providers. Human approval, budget controls, and quality gates remain part of the workflow.

Project overview

Orca is an agent development environment for coordinating multiple terminal-based coding agents across separate worktrees. It brings repositories, terminals, previews, agent state, account switching, and usage tracking into one workspace. A mobile companion can notify you when work finishes, show progress, and send follow-up instructions while the desktop sessions continue running.

Project overview

sandboxd is a self-hosted engine for building apps with coding agents. A request creates an isolated container, runs an agent against the prompt, and exposes the app at a live preview URL. Teams can use the browser console for creation and file review or integrate the API into their product; idle sandboxes sleep and wake on demand. It primarily targets Linux servers with Docker Compose and Git; macOS via Docker Desktop is best-effort. The project is a 0.x beta using containers rather than VMs, so untrusted multi-tenant deployments need additional hardening.

Project overview

Comet organizes coding work into guarded phases from specification through build, verification, and archive, recording state so an interrupted task can resume without reconstructing its history. It also helps create, compose, distribute, and evaluate reusable skills across many coding agents. The CLI requires Node.js 20 or newer.

Project overview

Golutra turns existing coding CLIs such as Claude Code, Codex, Gemini, and OpenCode into a shared multi-agent workspace. Users can run agents in parallel, orchestrate longer workflows, import or export templates, inspect individual logs, inject prompts during execution, and track handoffs in a visual interface. It keeps the underlying CLI tools and project layout intact, making it useful for developers who want centralized oversight without migrating their work to a new agent runtime.

Project overview

LobeHub treats agents as the units of work, allowing users to describe a role, create an agent, connect models, skills, and MCP-compatible plugins, and organize ongoing collaboration. It is designed for people who want one place to operate and review a persistent AI team rather than hand off tasks between isolated chats. The project remains under active development, and self-hosting requires a model API key.

Project overview

Hive turns an objective into a graph-based runtime that coordinates specialized agents in parallel while preserving role memory, checkpoints, and execution state. It adds human-in-the-loop controls, observability, cost limits, browser and compute tools, and LiteLLM-compatible hosted or local models; teams building only simple one-off chains may not need this heavier harness.

Project overview

SurfSense gives competitive-intelligence agents live connectors for social platforms, search, maps, and the open web through REST APIs or MCP. Scheduled and event-triggered agents can combine sources into cited briefs and alerts, then keep findings in a searchable knowledge base. The platform, including connectors, automations, and the MCP server, can be self-hosted with Docker, but the project states that it is still under active development and not yet production-ready.

Project overview

Run an LLM agent that can operate a real browser, terminal, files, keyboard and mouse, screen vision, and Android devices through a small set of local tools. GenericAgent records successful task paths as reusable skills, but you must supply a model API key and complete extra local setup for capabilities such as browser automation, vision, or computer control.

Project overview

MCK PPT Design Skill gives AI agents a consulting-style presentation system backed by a Python generation engine. Its workflow locks each slide's layout and takeaway in an outline, loads only the relevant design guidance, calls a library of slide and chart patterns, and runs machine-readable checks before and after rendering. It suits repeatable business-deck production and requires Python with python-pptx and supporting dependencies.

Project overview

cc-sdd installs a spec-driven development workflow as Agent Skills, moving from discovery, requirements, and design into task generation and long-running implementation. Each task can be handled by a fresh implementer using test-driven development, followed by an independent reviewer and an automatic debugging pass. The same seventeen skills target eight coding agents and thirteen languages; Claude Code and Codex are stable, while the other integrations remain beta.

Project overview

thClaws is an open-source agent harness that exposes a desktop GUI, CLI, headless mode, and web app from one Rust binary. It lets you switch among hosted or local model providers, attach MCP servers and skills, use persistent memory and knowledge bases, and coordinate scheduled jobs or agent teams without changing runtimes. Prebuilt releases cover macOS, Windows, and Linux. On headless Linux machines, the GUI build may require Wayland and WebKit libraries, while the CLI mode avoids those desktop dependencies.

Project overview

VibeSkills is a workflow runtime that takes one request, turns it into bounded stages, and binds the relevant locally installed skills for planning, implementation, testing, documentation, research, or review. It records verification evidence and preserves workspace context for later sessions. The runtime is invoked through the host application's Skills entrypoint rather than as a standalone CLI.

Project overview

Honeycomb gives coding agents such as Claude Code, Cursor, and Codex a shared memory that persists across sessions, tools, devices, and teammates. A long-running local daemon receives captures from hooks, CLI, MCP, and SDK clients, then exposes three levels of recall: compact keys, summaries, and full raw sessions. Storage is backed by Deeplake, so setup includes Deeplake authentication; the daemon is the sole storage client and centralizes those credentials behind loopback HTTP.

Project overview

arXiv Review Paper Harness is a portable Agent Skill for producing machine-learning and AI review papers with Codex or Claude Code. Starting from a topic, it organizes literature discovery, a human approval gate, issue-based drafting, citation-by-citation verification, prose refinement, and LaTeX compilation into an IEEEtran project with a PDF and BibTeX file. It requires Python, a working LaTeX toolchain, and web access for citation checks, and it does not begin writing the paper body until the plan is approved.

Project overview

TaskingAI is a backend platform for developing and deploying LLM-based agents. It brings models from multiple providers, tools, RAG systems, assistants, and conversation history under one console and unified API. Developers can prototype configurations in the UI, then integrate them into multi-tenant products through REST endpoints or the Python SDK. The community edition can be self-hosted with Docker Compose and requires configuration for the selected model services.

Project overview

VexJoy Agent provides a single `/do` or `$do` entry point that selects a domain agent and a matching workflow skill from a plain-language request. The routed task proceeds through explicit phases with test, review, and verification gates backed by hooks and deterministic scripts. The toolkit covers software work as well as research, writing, data analysis, moderation, and self-contained HTML artifacts.

Project overview

cera-reasoning-harness places an AI project's goals and instructions beside the chat, links them to each message, and saves the project locally so a later session can resume. It keeps a history of the interaction and its decisions; the project states that saved files are local and encrypted. The app targets Windows 10 or 11 and needs network access plus an AI service connection, so verify those privacy claims in the implementation before using it for sensitive work.