Solenta quickstart
Solenta runs every coding agent, claude, codex, opencode, kimi and grok , side by side in one native macOS window, each in its own PTY and git worktree. This is the short version: install it, meet the palette, and put a fleet to work. Solenta is in early beta; it updates itself, so you stay current.
Install
Download the beta disk image, open it, and drag Solenta to your Applications folder. It is signed and notarized, so Gatekeeper opens it without a fuss.
# from the Downloads folder
open Solenta.dmg
# drag Solenta → Applications, then launch it
- Requirements: macOS 14 (Sonoma) or later, Apple silicon or Intel.
- On first launch macOS may ask for microphone access, that is for the voice assistant, and you can skip it.
First launch
The first run walks you through a short setup wizard. You can skip any step and come back later from Settings. The steps are:
- Workspace, point Solenta at a project folder. A workspace maps one-to-one to a repo or directory.
- Agents, pick which coding agents you use. Anything not already on your machine can be installed with one click from the grid (claude, codex, opencode, kimi, grok). Hermes is different: it is an agent you provision on a VPS, not a local install (see Remote & Hermes).
- Voice, enable the voice assistant and choose a speaking voice.
- Tour, a quick look at the palette, orchestration, and the ambient (notch) layer.
The ⌘K palette
Everything starts with the launch palette. Press ⌘K anywhere to open Launch…, then pick an agent (or a plain shell), an optional task, and whether to run it in an isolated git worktree. The agents you use most sit as one-click chips in the top bar; everything else lives one keystroke away in the palette.
Workspaces & panes
A workspace is a project folder. Each workspace has tabs, and each tab splits into panes. Every pane is a real PTY running either an agent or a plain shell, arranged however the work demands.
- Isolated worktrees, an agent pane can run on its own git
worktree (
solenta/<slug>), so agents never touch your working tree or each other's. - Live status, hooks, plugins and notify shims report
running,blockedanddonefrom each agent, rolled up to every pane, tab and workspace dot. - In-split diff review, open a worktree's diff beside its pane and squash-merge it back as one commit, without leaving the split.
Voice assistant
Solenta has a built-in voice assistant with a speaking orb. Speech-to-text runs locally on-device (Parakeet TDT, accurate, 25 languages, a one-time model download), so push-to-talk works without shipping your audio anywhere.
- Talk: hold ⌥Space (or the mic button in the assistant), speak, and release.
- Setup: open Settings ▸ Voice to enable it, grant the microphone prompt, and pick a speaker. The speaker choice routes the on-device, cloud, and realtime-call voices.
- What it does: launches and steers agents by voice, and announces out loud when an agent finishes or blocks while you are away.
Orchestration
Instead of driving each agent yourself, launch an orchestrator: give it a goal and a set of worker profiles, and it fans the goal out into self-contained tasks, one worker agent per task in its own worktree. It reviews each worker's output, sends bounded rework until the work is up to par, and merges accepted work.
- Goal + workers: you supply the goal text and choose worker profiles from claude, codex, opencode, kimi and grok, and set a model for each.
- Reviewer and PR reviewer: optionally assign a dedicated reviewer agent to judge each worker's diff, and a PR reviewer that opens a pull request and reviews it before merge.
- Edit the plan: review and tweak the generated orchestrator prompt before you start the run.
- Runs smoothly: worker agents start ready to work without approval prompts, land beside the orchestrator in one tab, and finished workers close themselves once their work is merged.
- Confirm-first merges: a merge never happens on the model's say-so. When the orchestrator wants to merge, Solenta blocks on your explicit OK, approve, decline, or let it time out (which never merges).
Workflows alpha
Workflows is a visual canvas for designing multi-agent pipelines. Open it from the top-bar button, place agent nodes, connect them, and run the graph as a workflow: an orchestrator node fans out to workers, a reviewer checks their output, and results merge, all with the same worktree isolation and confirm-first merges as a plain orchestration run. It is early alpha and evolving quickly.
Notch overlay
On Macs with a display notch, Solenta drops a Dynamic-Island-style panel out of the notch on hover. It lists every agent with its workspace and status dot, so you can glance at the whole fleet, and jump to any of them , without switching windows. Blocked agents also raise a dock badge, a notification, and the bell.
Skills
Skills are short, agent-agnostic markdown playbooks Solenta can hand any agent, plain shell and git, no CLI-specific tool names. A few ship built-in:
- Code review checklist, review the diff before calling work
done; end with a verdict and
file:linefindings. - TDD: test first, write the failing test first, watch it fail for the right reason, verify green by exit code.
- Commit & PR hygiene, small logical commits, explicit paths, honest messages, root cause in the PR body.
Shared memory
Every pane connects to one knowledge graph, SQLite with FTS5 full-text
search, exposed over MCP. Before starting, an agent can call
agent_awareness for a digest of who is active, who is blocked,
and what the fleet already learned. Entries are superseded, never deleted,
so history stays auditable and the fleet remembers across sessions.
A once-a-day memory janitor agent (while the app is open) merges duplicate memories and tidies stale tasks; you can also run it on demand from the launch menu.
Checkpoints & rollback
Solenta auto-snapshots an agent's worktree as it works. The checkpoints button in a pane header (a clock icon) lists each snapshot with its relative time, file count, and subject. Click a row to restore the worktree to that point, and restore is itself undoable, because Solenta takes a safety checkpoint of the current state first.
Remote agents & Hermes
Agents don't have to run on your Mac. A background daemon
(agentmux-ptyd) can run on another machine, reached over an
SSH tunnel, and its agents keep running when you close the tunnel or quit
Solenta, the daemon lives on the remote.
- Any host: install the daemon on a machine you already have
(
scripts/remote-ptyd-install.sh user@host), add it under Settings ▸ Remotes, and launch agents on it from a remote workspace. - Hermes: a first-class remote agent that runs in a Docker VPS. The Add-Hermes wizard provisions the VPS, sets up keys, and connects , no manual server setup.
Auto-update
Solenta updates itself (via Sparkle). Because this is an early beta, updates land often; you don't need to re-download the disk image, the app checks for and installs new versions on its own. That is the whole point of the beta channel: you stay current automatically.