RunMaestro.ai @RunMaestroAI

Working in the TUI is equivalent to shackling your agents. Let them express themselves... not to mention the ability to create themed HTML dashboards.

keyboard_arrow_left Previous keyboard_arrow_right Next

@DottChen @charlieholtz @conductor_build Mobile is the missing piece for a lot of us. Maestro ships a QR mobile check-in today, so you can look in on your agents from your phone, and it works across Claude Code, Codex, OpenCode, and Droid. Open source and cross-platform. Give it a try.

@AdamBien We support Spec-Kit, OpenSpec, and BMAD natively in Maestro, our open source and cross-platform desktop orchestrator. Compatible with multiple providers today: Claude, Codex, Factory, Copilot, OpenCode with more coming. Give it a try.

@simas_ch We support Spec-Kit, OpenSpec, and BMAD natively in Maestro, our open source and cross-platform desktop orchestrator. Compatible with multiple providers today: Claude, Codex, Factory, Copilot, OpenCode with more coming. Give it a try.

A remote agent that blends into the local orchestrator is exactly the right shape. Maestro does remote agents over plain SSH, no relay service in the middle, so they show up right alongside your local ones in one desktop orchestrator across Claude Code, Codex, OpenCode, and Droid. Open source. Give it a try.

Thanks for sharing this. Agents messaging each other is something Maestro shipped months back: a group chat across sessions with a moderator AI that synthesizes agents holding different contexts, and it runs across providers too, Claude Code, Codex, OpenCode, Droid in one window. Open source. Worth a look.

@dsllwn @conductor_build Running more CLIs side by side is the right instinct. Maestro already does exactly that: Claude Code, Codex, OpenCode, and Droid in one window, no wrapper lock-in, plus a group chat where they can hand work between each other. Open source and cross-platform. Give it a try.

That failure mode, agents deciding they are done before they actually are, is the whole reason exit criteria matter. Nice work on /goal. Maestro Auto Run's Goal-Driven mode loops against explicit exit criteria with an iteration limit, so it either keeps going until the goal is genuinely met or caps the spend instead of becoming that expensive for-loop. Harness-agnostic too: any harness (Claude Code, Codex, OpenCode, Droid) and even local models. Open source, give it a try.

Loops with goal functions is precisely the shift. Maestro Auto Run is the desktop product built for it: Goal-Driven mode pursues an open-ended goal, looping until the exit criteria are met or it hits an iteration limit, so you really can set it running and go to sleep. It is harness-agnostic too, drive any harness (Claude Code, Codex, OpenCode, Droid) or even local models with the same loop. Open source. Worth a look.

That usage drain is the wrapper toll, you are paying for the tool on top of paying for Claude. Maestro does not charge for the wrapper: bring your existing Claude Code Max and Codex subscriptions and run them in one window. Free, open source, your machine, your keys, no credits or quota gating layered on top. Worth a look.

The "on my own machines" part is the whole game. Maestro is built that way by default: runs entirely on your machine, no account, no relay, open source. It is also multi-provider, Claude Code, Codex, and OpenCode side by side in one window, and remote boxes join over plain SSH if you want to spread the load. Worth a look.

The harness is the variable, not the model, that is the part people keep getting backwards. Maestro is built on that premise: Auto Run is a visual loop runner with a Goal-Driven mode that pursues a goal until your exit criteria are met, and it is harness-agnostic, so it drives any harness you point it at, Claude Code, Codex, OpenCode, even a lean local setup on Ollama. Not locked into Claude Code, not locked into one provider. Worth a look.

Give us a look see, you just described Maestro. It is a Desktop AI Orchestration Command Center: multi-provider group chat with a moderator AI, Cue event-pipelines, Ralph Wiggum GUI (Auto Run), remote agents over plain SSH, git worktree isolation, document graphs, Director's Notes, multiple Claude accounts at once, spec-framework native... the list just goes on and on. Open source, no relay, no account.

@SolutionsCay @superset_sh Add Maestro to that list if worktree management is the draw. Same shape, different posture: no account, no relay, open source, runs entirely on your machine. Git worktree isolation per agent, plus Claude Code, Codex, and OpenCode side by side in one window. Worth a look.

The churn is real, half the work lately is just re-rigging the setup every time the meta shifts. The fix is not picking the winner, it is a layer that does not care which one won: Maestro runs Claude Code, Codex, OpenCode, and local models side by side in one window, and a VPS box joins over plain SSH, no relay, so the Mac mini and the cloud agent can group-chat with each other. Swap the component, not the workflow. Worth a look.

The context-advantage point is the real one, the moat is not the model, it is the structured context you feed it. If your notes are the API, the relationships between them are the schema. Maestro has a document graph view that renders exactly that: your markdown wiki-links as an interactive graph, so you and the agent can see how the context connects.

@sauda_coder Plus Auto Run for the long unattended jobs. Open source and free. Worth a look.

17 free guides is a real contribution, that is the work that gets people past treating Claude Code like a black box. The wall after fluency is running several sessions at once without losing the thread, and that is the layer Maestro sits at: an orchestration GUI above the CLIs with parallel Claude Code sessions, even multiple Max accounts at once, and a moderator AI to synthesize across them.

@browomo That is exactly what Maestro Cue handles, event-driven pipelines (file changes, timers, agent-completions) that route work across the fleet automatically, with a group chat and moderator AI keeping them in sync. Open source and free. Give it a try.

12 agents running order ops for 140 buyers is the kind of operation that lives or dies on orchestration, not on any single agent. Past a couple of parallel agents the bottleneck stops being the model and becomes coordination: who runs when, who hands off to whom, what fires the next stage.