The automation platform that documents itself.
Runix crystallizes every workflow into a reusable .feature file. Generic drivers, autonomous AI agents, distributed swarm, scheduler, and live observability — one binary, one platform, one source of truth.
Quick start
recommendedcurl -fsSL https://runix.dev/install.sh | sh
runix.exe ai agent "take a screenshot and tell me what you see"
runix.exe swarm start --agents 2 --subscribe channel:*
Recent activity
System status
Real readouts from /api/runtime — driver source presence on disk, node_modules, and desktop-capture availability. This is the actual Pebble host you're talking to, not a mock.
Hand a real task to a real agent
Each card streams a live runix ai agent session. Output flows in the IDE console while the hosted engine works the task end-to-end.
The whole platform, one binary
Every capability below is a driver, a scheduler trigger, or a swarm role. Compose them in Gherkin or hand them to the AI.
⚙︎ Gherkin features
Drop a .feature file. Pattern-matched steps route to drivers in real time. Replay-ready. Reports auto-generate.
✦ Autonomous agent
One sentence in, reproducible workflow out. OpenAI · Anthropic · Gemini. Voice in / voice out. Resumable from checkpoint.
⌖ Distributed swarm
Ed25519-identified peers. Local file or NATS bus. Native gateway: Telegram · Slack · Discord · WhatsApp · iMessage.
⏱ Scheduler
7 trigger types: cron, interval, file watch, email IMAP, webhook, database CDC, SMS. Background daemon.
⌥ Vision & OCR
Custom vision model, OCR, and sharp-optimized screenshots for the agent loop. Pixel-accurate clicks.
⇄ MCP-native
Drop into VS Code Copilot, Claude Desktop, Cursor — any MCP host. Tools auto-register from every installed driver.
The IDE — drive the engine
Live editor with workflow syntax highlighting, workspace files, and a 6-mode runner: Feature, Task, CLI, Ask, Agent, Voice.
Talk to Runix directly
Fast AI Q&A plus one-click escalation into autonomous agent execution.
New chat
How can Runix help?
Ask anything or start from a quick prompt.
Workflows
Every workflow you have - yours, the examples, generated drafts - with what it needs and how it has been doing. Run any of them right here.
Triggers & schedules
Seven trigger types move work from manual launch to durable automation. Configure once, then let Runix keep it running.
Triggers
7 typesQuick presets
Active schedules
Runix Swarm
Tell the swarm what you want done. It figures out who to ask and how to split the work — you just watch the result.
.md board. Edit it here — the swarm picks up your changes automatically.Connect your computers
…Share the code. They type it on their computer — like joining a Discord server. Works for any number of computers.
Advanced · for Gods & Techno Priests
# general
▸ Advanced — spawn specific agents & send raw messages (only if you know what you're doing — most users should just use ▶ Run above)
Pick a specific role + driver allowlist. Spawning a planner here gives you the same auto-orchestrator the ▶ Run button uses. Other roles wait passively for messages. Use topic:<name> or channel:<name> as a Send target to broadcast or route through the gateway.
Spawn agent
Agent types & skills —
Agent types
Skills
Send message
Reach
Multi-machine: set
RUNIX_SWARM_TRANSPORT=nats + RUNIX_SWARM_NATS_URL on every node.Channels (Telegram/Slack/Discord/WhatsApp/iMessage) bridge through the gateway. Send to
channel:<name>.Full manual:
docs/user/SWARM_MANUAL.md in the Runix install.
Conversation tree —
Every message that flowed through the swarm — your task, the planner's decomposition, each delegated subtask, and the final answer. Subtasks show ↳ delegated · replies to X. Click any agent id to filter. Polls every 4s.
Audit timeline
Every typed event for a single trace. Paste a traceId (from the Messages tab, the engine log, or company-<id>) and walk the chain end-to-end. Persists to disk; survives restarts.
traceId above and click ▶ Load. Common trace ids:
company-<companyId>— every coordinator decision, delegation, budget charge, and HITL gate for that company.- The
correlationIdshown next to "↳ delegated · replies to X" badges in the Messages tab — walks one planner→executor handoff chain. - Any
traceIdfrom/api/swarm/events.
Execute as agent
Run a real autonomous agent task with the same tool-calling loop, screenshots, reports, and logs as the IDE's Agent mode. Output streams below; artifacts land in outputs/ and refresh in the strip.
Task & live console
no targetRun summary
—Recent run artifacts
Newest screenshots, reports, and logs from this host. Click to open. Updates after each run.
Modes
Who your agents are and how they work: instructions, the harness that runs them, what they may touch, their model and turn budget. Switch any agent's mode - a chat, a swarm agent, a running task - and it takes effect at its next step. Agents can switch their own and hand work to other modes.
Drivers & capabilities
Every driver is a peer WebSocket server. Drop a folder with a driver.json and Runix discovers it.
Action catalog
—Gherkin step catalog
—Everything Runix can do
The full capability matrix — 12 pillars, 31 categories, 450+ shipping features. Sourced live from /api/capabilities and /api/features; everything below corresponds to actual code in this repository.
Feature catalog
Filter by category or search by name/detail. Highlights matches inline.
Artifacts & runs
Browse every file the swarm produces — outputs userspace, agent workspaces, project boards, libraries (agents/skills), MCP queue, conversation scripts. Click any text file to preview & edit; the engine picks up your changes on the next iteration. Live screen moved to the global Preview dock — press Alt+P to see what's on screen from any view.
Recent runs
Voice & wake word
Whisper STT (90-95% accuracy) with native fallback. Wake word training via ElevenLabs. Continuous conversation mode.
TTS preview
Browser STT (live transcription)
Live speech recognition in Chrome and Edge; in other browsers Runix records your clip and transcribes it on the server. Click "Start" and speak - transcripts route to chat or the IDE.
Wake word recorder
Record three samples of your chosen wake phrase. Samples are kept locally (browser only) and the average duration is used as a heuristic threshold for client-side activation.
Voice interrupt commands
During engine voice mode (runix.exe ai agent --voice), these phrases trigger immediate state changes. The browser STT panel can also dispatch them when "Send transcripts to" is set to "IDE Agent task" and an agent is running.
🚪 quit — stop immediately
▶︎ go — resume
ℹ info — show progress
🔀 redirect — give new instructions
Marketplace
Everything you can add to Runix and its agents: drivers and integrations, modes, skills, instructions, prompts and collections.
Settings
Workspace, account, and engine configuration.
Account
Required for /api/ai-agent, /api/run-feature, /api/swarm/spawn, /api/swarm/send.
Build info
Theme
How Runix looks on this device. Also under View › Theme.
Startup
Runix has to be running to answer you. The wake word, schedules and the swarm node all stop when it does.
Connect your computers
Mission Control → one tap. No links to type unless you want Advanced.
How Runix works
Models, agents, voice, workflows and more. Saved to .env in your data folder, which updates and reinstalls leave alone.
Models in use now
Environment configuration
Live RUNIX_*, AI_*, OPENAI_*, ANTHROPIC_*, GOOGLE_*, GEMINI_*, ELEVENLABS_* + NODE_ENV/PATH/etc the engine sees. Secrets are masked. Anything Runix reads can be changed here, including variables not listed above.
Profile
Persisted to state/profile.json on the server. Drives default voice, model, and notification choices across the UI.
Privacy & compliance
Security controls
Audit trails, masked secrets, auth-gated APIs. A self-assessed control matrix tracks readiness against SOC 2 and ISO 27001; no audit or certification has been issued.
Multi-tenant
Tenant isolation, resource quotas, namespace-based security.
Self-hostable
One binary, your servers, your models. No cloud dependency.