Six steps from mailbox to memory.
Willder runs two agents — Research and Copy — connected by a shared memory graph. You connect your mailbox, review every draft, and the system gets sharper with every reply. Here is how each piece works.
- 01Connect
- 02Add contacts
- 03Start campaign
- 04Review drafts
- 05Read memory
- 06Connect your AI
Connect your mailbox
Go to Settings → Mailbox connections. Connect Gmail or Outlook in one click. Willder sends outreach from your real address — replies land in the same thread with native DMARC alignment. No shared sending pool.
Add your prospects
Add contacts and accounts manually or import a CSV. The research agent enriches each one automatically — company news, role context, recent signals — and writes findings to the shared memory graph before drafts run.
Start a campaign
Create a campaign with your ICP and tone. Toggle research mode on and the research agent runs first — searching the web, scraping sites, writing findings to memory — before handing off to the copy agent. Or draft straight from existing memory for speed.
Campaign
Q3 Outbound — Revenue Leaders
Prospects queued
18 contacts · 3 accounts
Run research first
Research agent enriches each prospect before drafting
Every draft needs your approval
No email leaves without your sign-off. Each draft lands in your inbox alongside the facts that shaped it. Read the context panel, edit the body if you want, then approve. Your edits are stored as training signals — the next campaign for the same account starts smarter.
To: priya@orrery.io · Draft #4471
Awaiting approvalPriya — saw Orrery just opened the Austin office and that you are hiring three AEs. Last time we spoke (Mar '25) ramp time was the open question. We cut it to 18 days for a team your size. Worth 20 minutes?
What we know
- Austin office opened · human
- Hiring 3 AEs · agent
- Cares about ramp time · extracted
Ask what you know
The memory graph grows with every enrichment run, every edit you make, every reply you receive. Go to Memory → Ask and query it in plain language. Upload a document, a call transcript, or a note and it becomes a queryable fact. The graph builds itself from your real history.
3 facts found
- Austin office opened Q2 '26
- Hiring 3 AEs, ramp time the blocker
- VP Revenue is Priya Anand
Connect your own AI
Issue a capability token in Settings → MCP. Paste one JSON block into Claude Desktop or Claude Code. Your AI now has memory_query, draft_list, draft_approve, and more — all through the same access gate, fully audited. Ask Claude what Willder knows without switching tabs.
Connection token — copy now, shown once
wcap_Xk9mR2vTq…CopyClaude Code — .mcp.json
{
"mcpServers": {
"willder": {
"command": "npm",
"args": ["run", "mcp"],
"env": {
"WILLDER_CAPABILITY_TOKEN": "wcap_..."
}
}
}
}A few things worth understanding
Can I use this without connecting a mailbox?
Yes — you can enrich contacts, build memory, and review drafts without a mailbox. You just can't send until you connect one. Good for building context before committing to outreach.
Does the research agent run every time?
No. Research is opt-in per campaign start. The default path is to draft straight from existing memory — fast and free. Enable research when you want fresh web signals before drafting.
Who can read the memory graph?
Access is scoped per user, role, and agent — default deny. Members get no access until an admin grants a scope. Every read is logged in the audit trail.
What happens when a contact replies?
The reply is matched to its outbound thread, stored as an inbound message, and written to memory as an extracted episode. The next draft for that contact starts knowing they replied.
What can the MCP connection do?
Query memory, fetch entities, write facts (write token), list and approve drafts, reject drafts, list campaigns. Everything goes through the same check() gate and is audited.
Can I edit the agents' prompts and tools?
Yes — go to Agents and hit Configure on any agent. You can set a custom system prompt, choose which tools the agent has access to, and pick the model. Custom agents can be created from scratch too.
Ready to run it.
Connect a mailbox, add your first ten contacts, and let the research agent run overnight. The first reviewed drafts land in your inbox in the morning.
Start free