The 30 Experts — pick, ground, ask, ship.
The default chat surface. 30 specialists across People, GTM, Ops, Data, Build, and Personal. They draft, audit, advise — they never send a real email or modify a real system on their own.
Pick an Expert
30 chat-driven specialists across six departments. Tap a card, the right system prompt loads and the right tools are wired in. You can switch Experts mid-chat; the new one inherits the conversation and the context. Experts are advisory — they draft, audit, advise. They never take real-world action on their own.
Drop the context
Four input modes in one chat: paste a URL (crawled in ~600 ms), upload a file (PDF · DOCX · MD · CSV up to 100 MB), connect a Drive folder, or paste text. The Expert always shows what it's grounded in.
Ask in plain English
US business English by default — clean, direct, no buzzwords. 120+ starter prompts across the Experts kill the blank page. Paste your last email and say 'continue this voice' to clone tone.
Memory carries
Every chat remembers the last one. The model learns whether you write short or long, whether you use 'we' or 'I', whether you cite numbers or stories. Turn it off per-chat if you need a fresh take.
Ship it
Edit inline, regenerate any block, save as a template. Export to Markdown · PDF · DOCX, or copy paste-ready HTML straight into Gmail, Confluence, Notion or whatever your team uses.
The 6 Agents — schedule, wire, watch, approve, hand off.
Agents go where Experts can't — they run on cron-style schedules, call real tools (ollagraph + memory + filesystem), write to a workspace you can read, and queue every outbound action for your approval before it leaves the platform. See /agents for the full list.
Pick an Agent
6 ⚡ Agents are different from the 30 ○ Experts. They run on a schedule (every 30 min, every 6 h) or on manual trigger. They call real tools — ollagraph scrapes, WHOIS, AEO audits — and write to a workspace you can read. Outbound SDR, SEO Auditor, Content Publisher, Inbox Triage, Pipeline Curator, Deep Research.
Wire the inputs
Each Agent reads from a workspace path: Pipeline Curator wants pipeline/deals.yaml; Inbox Triage scans inbox/unread/; SEO Auditor reads audit_targets.yaml. You drop the data, the Agent picks it up on the next run. Drag and drop from Drive or upload directly.
Watch the schedule
Each Agent has a built-in schedule. Outbound SDR every 30 min; SEO Auditor every 6 h; Inbox Triage every 15 min; Pipeline Curator every 4 h. Deep Research and Content Publisher are manual-trigger only. Pause any Agent with one click. See the next scheduled run, the last 10 runs, and the cost burned.
Approve before send
Action-taking Agents (Outbound SDR, Content Publisher, Inbox Triage) write to outbox/drafts/ and stop. You read the draft, edit it, click approve. Nothing leaves the platform without that click. Advisory Agents (SEO Auditor, Pipeline Curator, Deep Research) skip this — they only produce reports for you.
Hand off across Agents
Deep Research can drop a brief at handoffs/outbound-sdr/ and Outbound SDR picks it up. Content Publisher hands off to SEO Auditor for post-publish monitoring. Experts can hand off to Agents and vice-versa via shared memory namespaces. The 30 + 6 work as one workforce, not two silos.
Right model.
Every task.
All inference is routed through OpenRouter — never Anthropic. Experts mix OpenAI, Gemini, and Mistral based on the task. Agents pin specific models (Tongyi DeepResearch for the SEO and research workloads; GPT-4o-mini and GPT-4o for the rest). You don't pick. You don't tune.
Your files don't train anyone's model. Including ours.
India's Digital Personal Data Protection Act — designed-in, not bolted on.
Files and chats scoped to your workspace. Nobody else sees them.
Not our vendors'. Not ours. Inference only, per contract.
Hit delete and it's purged from storage and the LLM's logs within 30 days.