Open Orchestrators
Track the projects shaping multi-agent orchestration, then compare the builders, workspaces, company OSes, coordination protocols, and visual systems in one place. Closed products that matter to the community are separated from the open directory.
Latest news
Paperclip released v2026.427.0 with multi-user access, invite flows, structured issue-thread interactions, liveness recovery, sub-issue checklists, subtree pause/cancel/restore, and beta sandbox environments.
Orca's latest releases turn PR review into a first-class agent IDE flow: open a worktree from a pull request, ask an agent to review it, and send inline comments through gh, while v1.3.21 adds more PR and terminal setup polish.
Mastra's Apr 24 launch lets agents browse and act on websites, starting with Stagehand and AgentBrowser, while the linked @mastra/core release adds CLI-driven browser automation and live screencasts.
Directory
Open-source or publicly documented systems where orchestration is central to the product, sorted by GitHub stars when available.
Open-source personal AI assistant software built around chat interfaces, persistent context, skills, and execution.
Represents the assistant-shaped end of the orchestration space.
An agentic workflow builder combining workflows, chatflows, apps, and knowledge systems in one product surface.
Combines orchestration features with app-building and knowledge tools.
An MIT-licensed autonomous agent from Nous Research with persistent memory, self-created skills, scheduled automations, subagents, and messaging gateways.
Centers orchestration on a persistent personal agent that learns across sessions, delegates parallel work, runs scheduled tasks, and reaches operators through CLI and chat platforms.
Open-source orchestration for zero-human companies, framed around AI employees, goals, and jobs across business functions.
Frames orchestration at the company and business-function level instead of only task flow.
An open-source system for building AI agents visually, with emphasis on agent flows and orchestration through a builder interface.
Presents orchestration through a visual builder rather than a code-first framework.
Organizes work as crews of specialized agents with a role-based model for delegation, collaboration, and task structure.
Makes the team metaphor central to the orchestration model.
An open-source runtime for building, running, governing, and monitoring agents, teams, workflows, and AgentOS services in production.
Centers orchestration on production agent runtimes, teams, workflows, memory, guardrails, and AgentOS operations.
An MIT-licensed personal AI assistant that runs Claude agents in isolated containers, connects to chat channels, keeps memory, schedules work, and uses skills as git branches.
Centers orchestration on a small, understandable OpenClaw alternative with OS-level container isolation and multi-channel personal-agent operation.
An open-source AI agent platform for building agents with integrations, workflows, knowledge bases, tables, and docs.
Leans toward a workspace-style product rather than a narrow framework.
A workflow layer for OpenAI Codex CLI that adds stronger default sessions, reusable skills, native hooks, HUD/status surfaces, project guidance, and team-style execution commands.
Centers orchestration on making Codex a repeatable agent runtime: `$deep-interview` for clarification, `$ralplan` for planning, `$team` for coordinated parallel execution, and `$ralph` for persistent completion loops.
An open-source Kanban workspace for planning, running, reviewing, and shipping parallel coding-agent work.
Centers orchestration on issue planning, coding-agent workspaces, branches, terminals, dev servers, diff review, browser preview, inline feedback, PR creation, and merge flow.
An open-source managed agents platform where coding agents act like teammates, take assigned issues, execute autonomously, and reuse shared skills over time.
Centers orchestration on issue boards, runtime daemons, and reusable skills for human-plus-agent teams.
An open-source TypeScript framework for building agents, graph-based workflows, MCP servers, evals, observability, and production AI applications.
Centers orchestration on TypeScript agents, explicit workflows, human-in-the-loop suspension, MCP, evals, and observability.
A multi-agent workspace manager for Claude Code, GitHub Copilot, Codex, Gemini, and other coding agents with persistent work tracking.
Centers orchestration on git-backed work state, handoffs, and workspace-level coordination for coding agents.
A local-first agent workspace and host layer for teams, with desktop and CLI runtimes, shareable setups, browser control, scheduled tasks, and remote operation.
Centers orchestration on shareable agent setups: local or remote hosts, permissions, browser automation, scheduled prompts, and opencode-powered execution.
A local code editor and control plane for running many CLI coding agents in parallel across isolated git worktrees.
Centers orchestration on parallel coding-agent execution, worktree isolation, monitoring, review, and editor handoff.
A TypeScript library and CLI for orchestrating AI coding agents in isolated sandbox environments with branch strategies, hooks, logs, templates, and merge-back workflows.
Centers orchestration on programmatic sandboxed coding-agent runs: invoke an agent with `sandcastle.run()`, run it in Docker, Podman, Vercel, Daytona, or a custom provider, and bring the resulting commits back through a branch strategy.
An open-source GitHub Copilot-based agent team system where specialists live in the repo, keep memory, route work, and run in parallel.
Alpha software; centers orchestration on repo-native team files, coordinator routing, persistent memory, parallel execution, watch mode, and CLI/SDK workflows.
A worktree-based IDE for AI coding agents that runs Claude Code, Codex, OpenCode, and other agents side by side across isolated git branches.
Centers orchestration on parallel coding agents, worktree isolation, and in-app review flow inside a desktop IDE.
An AI-first knowledge base where files live on disk and a team of AI agents help with execution.
Combines files, knowledge, and agent execution in one operating surface.
A self-hosted multi-agent AI runtime and control plane for autonomous agents, delegated work, schedules, provider management, MCP support, and chat-platform connectors.
Centers orchestration on a self-hosted runtime where agents can delegate work, manage providers, use connectors, run scheduled tasks, and operate OpenClaw and Hermes Agent swarms.
An Apache-2.0 office suite where AI agents collaborate with humans on docs, databases, slides, and flowcharts through MCP and comments.
Centers orchestration on shared office artifacts: agents can be mentioned, receive contextual tasks, edit content, leave traceable changes, and work through MCP without replacing the user's existing agent.
A governance-first AI agent orchestration platform where policy, approval, budget, classification, and audit checks run before agent actions execute.
Centers orchestration on structural governance: pre-action enforcement, divisions, tiers, budgets, audit trails, always-on agents, and self-hosted operation.
A coordination-oriented system with a protocol, rooms, agent lifecycle docs, a CLI, and multiple agent clients.
Structured around agent coordination, communication, and lifecycle management.
Not open, important
These products are not open-source projects, but they are important enough to track because they influence how builders think about multi-agent orchestration.
Augment's answer to what comes after the IDE: agent orchestration, isolated workspaces, and living specs in one developer workspace.
Not open-source; tracked as an important category signal because Intent puts coordinated agents, isolated workspaces, and living specs in one developer workspace.
A closed AI-agent platform for performance marketing where agents connect to the marketing stack, analyze data, launch campaigns, generate reports, and run scheduled workflows.
Not open-source; tracked as an important category signal because Hyper turns marketing work into an agent operating layer with tool connections, custom agents, action controls, reporting, and MCP-aware integrations.
A closed autonomous company OS for running daily operations with a self-improving multi-agent growth team.
Not open-source; tracked as an important category signal because Crewlet frames multi-agent orchestration as an operating layer for growth, engineering, support, and data work across existing company tools.
Governance
Tools that sit between agent intent and execution: policy checks, approval gates, budgets, permissions, and audit trails.
MIT-licensed runtime governance toolkit for AI agents with policy enforcement, identity, sandboxing, reliability, and compliance checks.
Sits between agent frameworks and actions so tool calls, resource access, and inter-agent messages can be evaluated before execution. Microsoft documents Python, TypeScript, .NET, Rust, and Go SDKs, plus integrations for LangChain, CrewAI, AutoGen, OpenAI Agents SDK, Google ADK, Semantic Kernel, Dify, LlamaIndex, Haystack, LangGraph, and related stacks.
Open governance toolOpen-source Python SDK for pre-execution governance gates around AI agents, backed by Postgres.
Adds scope checks, budget caps, human approval tokens, loop detection, agent presence and halt handling, behavioral contracts, and HMAC-chained audit trails for actions routed through the SDK. Tracked as agent-friendly tooling, not an orchestrator runtime.
Open governance toolOpen-source authorization layer that intercepts AI agent tool calls before execution and evaluates them against policy.
Provides TypeScript and Python SDKs, YAML rules, local deterministic evaluation, optional semantic checks, human approval routing, audit logs, and integrations for OpenAI, Anthropic, LangGraph, CrewAI, MCP, browser agents, and custom tool-calling setups. The core repo is Apache-2.0; Veto Cloud adds managed commercial services.
Open governance toolCommercial Databricks platform layer that extends Unity Catalog governance to LLM access, MCP servers, APIs, and agentic AI workflows.
Tracks LLM and MCP calls, controls which agents can access external systems with fine-grained permissions and on-behalf-of access, and adds observability, cost attribution, guardrails, fallback models, and rate limits. Included as a commercial governance layer, not an open-source toolkit.
Open governance toolTrust and policy enforcement gateway for agent-to-agent settlement, reputation, spending limits, provenance, escrow, and audit trails.
Sits between agents and an A2A Settlement Exchange so settlement requests can be governed by reputation thresholds, spend caps, provenance requirements, health checks, alerts, and cryptographic audit logs. The related A2A Settlement repo is MIT-licensed; SettleBridge's public license metadata is inconsistent across the website, README badge, and GitHub detection, so it is tracked as a trust governance layer rather than a clean OSS claim.
Open governance toolTools
Web analytics for builders that Claude Code, Codex, Cursor, OpenClaw, Paperclip, and similar AI agents can use.
Exposes traffic, usage, sources, funnels, conversions, projects, stats, and experiments through plugin, skill, MCP, CLI/npx, and API so agents can set up tracking, verify events, brief you, find conversion leaks, and run growth experiments without a dashboard-first workflow.
Open toolReusable companies for AI agents: pre-built organizations that can be installed with a single command.
Useful when orchestration needs reusable agent-company structures instead of starting every multi-agent team from scratch.
Open toolObservability for OpenClaw agents that shows what failed, where spend leaked, and how to improve runs.
Adds traces, trajectory dashboards, token and cost visibility, and Tracy, a doctor agent that recommends fixes from live run data.
Open toolLocal TUI observability for AI coding-agent sessions, tokens, cost, tool failures, latency, anomalies, diffs, and CI evidence.
Parses local logs from Claude Code, Codex CLI, Gemini CLI, Aider, Cursor exports, OpenClaw, and related coding agents so builders can inspect session health without sending prompts or code to a hosted tracing service.
Open toolmacOS workspace for running Claude Code, Codex, Gemini CLI, and other agentic CLIs in parallel sessions.
Runs each agentic CLI as a real PTY in the user's shell, with project boards, auto-created git worktrees, session resume, git diffs, file editing, labels, dependencies, and quick commands. Tracked as a tool/session workspace, not an orchestrator runtime.
Open toolServices
Services companies and consultants that publish practical material on deploying, hardening, and operating multi-agent systems. These are not player entries.
Malta-based AI innovation group helping organizations understand, implement, and scale AI through consulting, development, and products.
Tracked as an implementation services entry, not an orchestrator. Nomad Inno publishes services around AI strategy, responsible AI, custom LLM development, fine-tuning, RAG, system integration, process mapping, workflow automation, and practical AI implementation through Artificial Impact, Peachstone AI, and Systems Impact.
Open serviceImplementation services and practical writing for teams moving AI agents from pilots into production operations.
Tracked as an implementation services entry, not an orchestrator. Public contact details list United States and Canadian offices in New York and Montreal; no explicit worldwide coverage claim is made. Relevant because Agentix publishes production-agent guidance around OpenClaw implementation, observability, run reviews, security reviews, permissions, evals, guardrails, and operating models.
Open serviceContribute
To add a project, fix a link, update a description, or suggest a tighter comparison, open a pull request or issue in the repo.
Repository: github.com/Agent-Analytics/awesome-multi-agent-orchestrators
What a good update looks like
