Shared customer memory for B2B SaaS
The shared customer memory for B2B SaaS. So your team and their AI agents get answers without digging through tools.
ask recon
1 of 6Brief me on Acme before my call tomorrow.
Enterprise plan, 52 seats, 17 active this month. Two open tickets (search regression, kanban bug) filed by VP Ops. Renews May 6.
Product overview
A short walkthrough of what Recon reads, the shared customer memory it builds, and how monitors and advisories turn that memory into the next action.
How it works
Memory under Accounts. Watch under Missions. Action under Advisories.
Plan, usage, recent issues, admins, opportunities — one cited file per account.
Recon watches the accounts that matter. Pings you when usage drops, tickets land, or renewals slip.
When something is worth doing, Recon proposes the action with cited evidence. You approve before anything writes.
One shared memory. Recon remembers every customer, watches the accounts you care about, and proposes the next move with evidence attached.
Every customer gets a cited history: plan, usage, admins, incidents, tickets, billing, adoption, risks, and opportunities. Ask questions when you need to, but the memory is already there.
Every SQL query run, every file read, every ticket checked — visible and auditable. No black boxes.
Postgres, Supabase, GitHub, Linear, Intercom, Zendesk, HappyFox, Sentry, Stripe, HubSpot, and docs. Recon reads the systems where customer truth actually lives.
Save Recon's answers as a Notion page, a Linear ticket, a Slack thread, or a downloadable file. No copy-pasting across tools.
Upload docs, notes, PDFs, or call transcripts when they matter. Recon folds them into customer memory and cites them in answers.
Define what matters in plain English. Recon keeps an eye on specific customers, usage drops, renewals, issues, and spikes.
Daily synthesis reads across monitors, investigations, and account history to update health, risks, and opportunities.
Actionable recommendations with evidence attached. Create a ticket, investigate further, start a monitor, or prep outreach after approval.
RoadmapBuilding an MCP server so Claude Code, Cursor, and ChatGPT can read your customer memory.
Security and trust
Recon reads your database, code, errors, tickets, and billing to remember every customer. That access is enforced read-only, gated by approval for any write, and logged end to end.
Read the full security overviewDatabase access through a dedicated read-only Postgres role. Writes require human approval.
Every claim links to its source. Every query and tool call lands in an append-only log.
Each investigation runs in an ephemeral E2B Firecracker microVM, torn down on idle.
Connection credentials encrypted at rest. End users never see raw keys.
Your Anthropic, OpenAI, or Google key. Recon never proxies tokens through a shared account.
No per-seat charges. Unlimited users via Slack. Bring your own API key for inference, so tokens are billed by your provider, not us.
For one team
For multiple teams
Every plan includes a guided setup call with our team.
Anything you ask Recon. An investigation, a monitor run, or a follow-up question all count as one conversation, regardless of how many steps Recon takes to answer. Failed conversations (sandbox errors, integration auth issues) do not count.
One call, your systems connected, every customer remembered.
Most teams are live in under 30 minutes.
Curious? Ask an AI about Recon