Incident investigation
Log reader + deploy historian + metrics analyst + ticket triager — all parallel, surfacing the pattern a human would miss.
Parallel workEarly access
Every question answered in parallel.
A lead agent breaks your request into focused pieces and delegates each to a specialist — deploy logs, metrics, tickets, code. They work at the same time. You get one synthesized answer.
No card · one-click signup · unsubscribe anytime
The problem
Real business questions span tools, people, and timelines. Sequential tool-use turns a two-minute investigation into a ten-minute wait. Parallel specialists let one question spread across every source at once, then come back synthesized — faster than a human could open the tabs.
How it works
Four steps from zero to running in production.
- Step 1
Plan
A lead agent reads your question and breaks it into focused subtasks — one per tool, domain, or data source.
- Step 2
Fan out
Each subtask is handed to a specialist with its own model, prompt, and toolset. Up to 25 run in parallel.
- Step 3
Synthesise
Specialists deliver their findings back to the lead. It merges, resolves conflicts, and drafts the final answer.
- Step 4
Trace
Every step is visible on the timeline. See which specialist touched which tool, what they concluded, and why.
What you get
Built for the way teams actually ship work.
- Up to 25 concurrent specialist threads per session
- Per-specialist model selection — fast coordinator, deep subagents
- Shared filesystem so specialists can hand off intermediate work
- Tool approvals still route through one human-in-the-loop inbox
- Full execution trace exportable for audit or review
- Works with Company Brain and auto-evaluation without extra setup
Real scenarios
Where it earns its keep.
Deal research
CRM specialist + news crawler + product-usage analyst — one fan-out builds the account brief.
Release prep
Changelog writer + doc updater + PR reviewer — fanned out, synthesised into a release package.
Customer onboarding
Policy advisor + config checker + integration helper — one specialist per workflow, parallel to a single customer Q&A.
Early signal
What the numbers say.
Benchmarks from the underlying platform research and early-customer pilots. Your mileage will vary with scope and setup.
25
Max parallel specialists
< 30s
Typical investigation
fraction of Opus
Cost vs. serial
Frequently asked
Things people ask us.
Does this cost more than a single agent?
Often less. The coordinator runs on a cheap fast model (Haiku); specialists only escalate to bigger models when they genuinely need to. Overall cost is usually lower than serial Opus calls.
Who approves the specialists' tool calls?
You do. All approval requests fan into one inbox on the session timeline, so even a 25-specialist investigation feels like a single conversation to the human in the loop.
Can we author custom specialists?
Yes — that's on the roadmap for the Business tier. Admins will define a specialist's name, system prompt, model, and tool access, then publish it to the roster.
What happens if one specialist fails?
The coordinator sees the failure on the shared event stream and either retries with a different specialist or returns a partial answer with the gap called out.
Get early access
Be first on One ask, a team of specialists.
We’re onboarding a small group before general release. Tell us a bit about your team and we’ll reach out when the next slot opens.
Keep exploring
Vertical Knowledge Packs
Support, legal, finance, incident response. Pre-curated knowledge and rubrics that make your agent useful from day one.