Agents
Six agents on day one. Build the rest yourself.
Every agent — built-in or custom — runs on the same engine: durable tasks, persistent memory, append-only logs, and approval gates on everything outbound.
Workspace Agent
Live with Google WorkspaceYour inbox and calendar, one agent. It triages email, drafts replies in your voice, reads real availability, and books meetings that respect your working hours. Nothing sends without your approval.
[workspace-agent] triage 23 → 2 drafts · kickoff booked thu 10:00
Social Media Manager
Built-inDrafts platform-appropriate posts in your brand voice and builds content calendars. Every post waits at an approval gate before it goes anywhere.
[social-agent] 3 drafts → awaiting approval
Lead Generation Agent
Built-inEnriches inbound leads from context and memory, scores them 0–100 on fit and intent, and routes hot ones to you with a recommended next action.
[lead-agent] scored 87 → recommended: reply today
Routing Agent
Built-inClassifies incoming work and dispatches it to the agent best equipped to handle it — the traffic controller of your constellation.
[routing-agent] classified → dispatch:workspace
Workflow Analyst
Built-inStudies how work moves through your organization and proposes repeatable workflows — with human approval gates built into every one.
[analyst] pattern found → proposed workflow · 5 steps
Oracle
Live · Deep ResearchAsks Claude, GPT, Gemini, and Grok the same question — each grounded in live web search — then a reasoning model reconciles them into one cited consensus, with disagreements surfaced, never hidden.
[oracle] 4 models → consensus 92% · 8 sources
Deep Research
Four minds. One cross-checked answer.
One question goes to four frontier models in parallel — Claude, GPT, Gemini, and Grok, each grounded in live web search. A reasoning model reconciles their answers into a single consensus with a confidence score, explicit disagreements, and citations.
Diverge
Your question fans out to four frontier models at once — each researching independently with live web search, blind to the others.
Converge
A reasoning model reads all four answers and reconciles them into one consensus, scored with an explicit confidence level.
Transparent
Where the models agree, you see it. Where they disagree, each position is mapped — with sources. No black-box averaging.
[oracle] 4/4 answered → consensus 92% · 2 disagreements mapped · 8 sources
Agent Builder
Describe the agent. Arcasya builds it.
01
Describe it in plain language
“An agent that watches inbound invoices and flags anything unusual.” That's the whole spec.
02
ArcasyaAI generates the agent
Persona, system prompt, tools, triggers, and a step template — generated and validated against the tool registry.
03
It joins your constellation
Installed into your organization immediately. A new orb appears in Orbit and it's ready for goals.
Custom agents aren't second-class. They run the exact same engine as the built-in six — same durable queue, same memory, same approval gates, same execution log.
Deploy your first agent in minutes.
Start free and dispatch a real goal to a real agent today.