Try for free

Ultron Docs

Welcome to
Ultron Documentation

Report a bug
Status

Try a skill.

Pick any of the six below and you land inside the app with the whole workflow already loaded, briefed, and waiting for your first command.

Skills· slash them in chat

Competitive analysis

Pull rivals, score positioning, surface gaps.

Run it in the app

From command to production.

Nothing below is mocked for the page, because the commits are actual production deploys, the plan is a live run mid-flight, and the scaffold is exactly what the CLI hands you when a new skill is born.

Recent commits· main
05-12Ultron joins your calls - Recall.ai bot drops in, takes structured notes, and writes a striker-post-call deliverable to your CRMcalls
05-12Computer is stable - persistent E2B sandbox per workspace with a VNC pipeline so the agent can drive a real browser end-to-endcomputer
05-12Background jobs center - donut indicator under the chatbox plus a full panel that tails every running and recent jobjobs
05-08502 → sandbox_dead, start.sh boot resilience, booting state, safer reconcilersandbox
05-08add quote + stat_card primitives + 2 new layout templatescarousel

What shipped, when.

Every entry is a real production deploy. Subjects come straight from the commits, no rewrites.

View full changelog
Active plan· 3 steps
pull rivals + score positioning2.1s
draft outreach sequencerunning
sync to CRM + notify HITLqueued

Plan, run, observe, iterate.

Ask in plain English. Ultron drafts a plan, runs the skills, and pauses at every decision that leaves the building.

Workflow runtime
> /cold
/competitive-analysisresearch
/cold-outreachsales
/morning-briefingops
/vc-prospectorlead-gen

Run a skill.

Trigger any of the 57 skills from chat with a slash command or plain language.

View slash commands
quarterly-recap· research
skill.yaml192 B
prompt.md1.4 KB
tools.allow86 B
registry reload pending

Build a skill.

Scaffold a skill with a manifest, a tool allow list, and a model tier. Loaded on next start.

Skill anatomy

Skills, techniques, and field notes.

Everything deeper lives one click from here, from the full skill catalog to the hand-crafted workflows and the teardowns we wrote after running real campaigns.

Skill catalogEvery skill by category, with its manifest, tools, and model tier.TechniquesReusable, hand-crafted workflows you can run end to end.ResourcesField-tested teardowns, straight from real campaigns.ArchitectureHow chat, skills, memory, sandbox, canvas, and swarm fit together.

Every frontier model, one runtime.

Whichever lab ships the best frontier model this month, it is already wired into the same runtime, so you can pin one per skill or let the router decide on every turn.

claude-opus-4-8
claude-sonnet-4-6
gpt-5.5
DeepSeek-V4-Pro
model-router
Kimi-K2.6
grok-4-20-reasoning
FW-MiniMax-M2.5Fireworks
qwen-qwen3.5Hugging Face

How they stack up.

Every model runs the same benchmark, where higher quality and throughput win the column and lower violation rate and cost win theirs.

ModelQuality index
  • 01gpt-5.50.82
  • 02claude-opus-4-60.82
  • 03gpt-5.40.81
  • 04claude-opus-4-50.81
  • 05Kimi-K2.60.79
  • 06DeepSeek-V4-Pro0.78

Indicative benchmark figures. Accent marks the best result in each column.

Six systems that make it run.

Every turn you run leans on the same quiet machinery underneath, six subsystems passing work down one line from the skill you invoke to the background job that quietly finishes it.

Skills57 skills across 8 categories. Composable, registry based, scoped tools and model tier per skill.
MemoryThree storage surfaces, hybrid retrieval, and a four step conversation compression cascade.
WorkspacePersistent sandboxes with a live remote desktop, bundled engineering skills, and a 60 second reconciler.
CanvasRuntime renderer with 30+ block types. Export to PNG, PDF, or save to Brain.
Tools91 native tools plus external servers. Provider modules with status labels and HITL approvals.
Background jobsReconcilers, janitors, lifecycle emails. The quiet machinery that keeps state honest.