{"slug": "muxlio-durable-agent-workflows", "title": "Muxlio – Durable Agent Workflows", "summary": "Muxlio released an Apache-2.0-licensed tool that runs multi-step workflows across AI agents and command tools, saving progress so a client can disconnect while the coordinator keeps running. The tool installs via Homebrew on macOS or Linux and bundles a CLI, MCP server, adapters and web UI; its example \"implement-features\" recipe runs Implement → Review → Run checks → Commit locally → Next feature, pausing for attention after ten unsuccessful verification attempts. Muxlio's author says he uses it to develop Muxlio itself, running it on a Mac mini and queuing work through an assistant.", "body_md": "Build a workflow. Let your agents run it.\n\n[Documentation](https://muxlio.github.io/muxlio/) ·\n  [Get started](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/service.md) ·\n  [Implement-features recipe](https://github.com/Muxlio/muxlio/blob/main/recipes/implement-features/workflow.ts) ·\n  [Apache-2.0](https://github.com/Muxlio/muxlio/blob/main/LICENSE)\n\nMuxlio runs multi-step workflows across AI agents and command tools. You or your assistant define the steps, their inputs, and the checks they must pass. Muxlio executes the workflow, saves progress, and gives you a place to follow the work.\n\nUse it for work that needs several steps, checks, or hand-offs. Your client can disconnect while the coordinator continues running. You can return to the run's status, logs, outputs, and any failure that needs attention.\n\nI use Muxlio to develop Muxlio itself. The\n[implement-features recipe](https://github.com/Muxlio/muxlio/blob/main/recipes/implement-features/workflow.ts) works through\nfeatures one at a time:\n\n```\nImplement → Review → Run checks → Commit locally → Next feature\n                         │\n                         └─ if checks fail: Fix → Run checks again\n```\n\nThe workflow runs the repository's build and tests before committing. After ten unsuccessful verification attempts it pauses for attention. You can inspect the change, review notes, test output, and execution history before deciding to push. Each verified commit becomes the starting point for the next feature.\n\nTo run the batch in a private Git clone, follow the\n[Docker Sandbox walkthrough](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/docker-sandbox.md). One sandbox worker\noffers Claude Code, Codex, Cursor, and shell commands. The example takes two\ndependent features from setup through preflight, execution, and fetching the\ncommits. The [live recipe](https://github.com/Muxlio/muxlio/blob/main/recipes/README.md#append-features-to-a-running-batch)\nalso lets you add features to the same branch while the batch is running.\nIf you set up both workspaces, choose `localhost` to run on the host or `sandbox`\nto run on the `sbx` worker.\n\nI keep Muxlio running on a Mac mini, queue work through my assistant, and review progress later. You choose the machine: your workstation, another computer, or a server. The execution host needs to stay running; your chat session does not.\n\nThis is one workflow, not a special product command. Adapt the\n[recipe's tasks and checks](https://github.com/Muxlio/muxlio/blob/main/recipes/implement-features/tasks.ts) to your own project,\nor build a workflow for research, data processing, documents, or other work.\n\nWith [Homebrew](https://brew.sh) available on macOS or Linux:\n\n```\nbrew tap Muxlio/muxlio https://github.com/Muxlio/muxlio\nbrew install --HEAD Muxlio/muxlio/muxlio\nmuxlio version\n```\n\nThis builds the current `main` branch and installs the CLI, MCP server, adapters,\nand web UI. Homebrew handles the build dependencies; you do not clone the repo.\nInstall and sign in to the agent CLIs your workflows will use separately.\n\nWith Muxlio installed, initialize the workspace where work should happen:\n\n```\nmuxlio init --client kiro --scope project\nmuxlio service start\n```\n\nChoose `--client` for your authoring assistant. `init` allows all bundled\nexecutors by default, including agent tools, commands, workspace writes, and network access. Change\n`executors` in `~/.muxlio/config.json` to restrict permissions; repeated init\npreserves explicit policies and disabled executors. `init` prepares the workspace settings,\ncredentials, agent instructions, and MCP settings. The service starts the\ncoordinator, local worker, UI, and configured MCP listener together.\n\n[Connect your assistant through MCP](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/mcp.md), then ask it to build a\nworkflow for your task. Describe the inputs, the steps to run, and the output you need.\nHave it validate the workflow and show you the plan before starting the run.\nDevelopers can also author workflows directly with the TypeScript SDK.\n\nThe installed service does not need Muxlio's source checkout. Workers need the\nfiles, authenticated agent CLIs, and tools required by your workflow.\n[Source-build instructions](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/developer-reference.md#build-from-source)\nare available if you prefer to build the runtime yourself.\n\nOpen **[http://localhost:7600](http://localhost:7600)** or ask your assistant for run status. Inspect the\nexecution graph, attempts, live logs, and generated artifacts. Pause, resume, or\nretry when appropriate. Uncertain external effects require reconciliation before\nwork can safely repeat.\n\nTo launch a registered workflow from the dashboard, select its version and an\nauthorized workspace, enter the inputs, and run preflight. See the\n[dashboard guide](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/ui-operations.md#launch-a-workflow).\n\n```\nmuxlio service stop\nmuxlio service restart\n```\n\n| Start here | Go deeper | \n|---|---|\n| [Which adapter should I use?](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/adapters.md) — native CLIs, ACP, and A2A | [Adapter capability reference](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/developer-reference.md#run-with-real-adapters) | \n| [Service Guide](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/service.md) — setup and service controls | [Service reference](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/service-reference.md) | \n| [Docker Sandbox walkthrough](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/docker-sandbox.md) — run dependent features in a private clone | [Worker protocol](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/worker-protocol.md) | \n| [MCP Integration](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/mcp.md) — build and run from your assistant | [MCP tools and protocol](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/mcp-reference.md) | \n| [Observability UI](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/ui-operations.md) — follow progress and results | [Operations API](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/operations-reference.md) | \n| [Benchmarks](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/benchmarks.md) — what the suite checks | [Benchmark methodology](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/benchmark-methodology.md) | \n\nExplore the [architecture](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/overview.md),\n[execution and security contracts](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/execution-security.md),\nand [CLI, SDK, and configuration reference](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/architecture/developer-reference.md).\nThe [example recipe](https://github.com/Muxlio/muxlio/blob/main/recipes/implement-features/workflow.ts) and\n[authoring guide](https://github.com/Muxlio/muxlio/blob/main/muxlio-skills/SKILL.md) show how workflows are put together.\n\n`muxlio init --client NAME --scope project --workspace /absolute/project` installs\nthe discoverable Muxlio skill and its references in that project. Use `--scope user`\nfor discovery across projects. See [client skill setup and invocation](https://github.com/Muxlio/muxlio/blob/main/muxlio-docs/mcp.md#install-and-invoke-the-muxlio-skill)\nfor supported paths, Cursor version handling, and preservation of existing edits.\n\nInstalled releases support offline JSON authoring: run\n`muxlio template echo > workflow.json`, then `muxlio check workflow.json`.\nThe `echo` template needs no executor; `muxlio template task` shows a generic\ntask with input and output bindings. Templates are embedded in the binaries\nand shipped in `share/muxlio/templates`. With MCP, use `workflow_template`,\nthen `check_plan` and `start_run`. No Node packages or registry access are\nneeded for JSON authoring, validation, fake testing, or submission.\n\nSee [the installed authoring guide](https://github.com/Muxlio/muxlio/blob/main/muxlio-skills/references/authoring.md) for\nan empty-workspace example and submission instructions. Optional TypeScript/\nJavaScript compilation executes arbitrary code, requires\nproject-local dependencies installed from\nyour pinned lockfile, and never implicitly downloads packages.", "url": "https://wpnews.pro/news/muxlio-durable-agent-workflows", "canonical_source": "https://github.com/Muxlio/muxlio", "published_at": "2026-11-19 21:32:09+00:00", "updated_at": "2026-11-19 21:44:45.510579+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "agent-protocols"], "entities": ["Muxlio", "Claude Code", "Codex", "Cursor", "Homebrew", "Kiro", "MCP"], "also_reported_by": [], "takeaway": "Muxlio's implement-features recipe runs build and tests before each local commit and pauses after ten failed verification attempts, letting a developer review changes and test output before pushing.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/muxlio-durable-agent-workflows", "markdown": "https://wpnews.pro/news/muxlio-durable-agent-workflows.md", "text": "https://wpnews.pro/news/muxlio-durable-agent-workflows.txt", "jsonld": "https://wpnews.pro/news/muxlio-durable-agent-workflows.jsonld"}}