Now accepting design-partner engagements
An AI deputy for every customer-engagement.
Outpost spawns a persistent deputy per customer-engagement so one Forward-Deployed Engineer can carry more engagements at consistent quality. Drafts Friday status updates, triages eval failures, keeps a paper trail.
Built for Heads of Forward-Deployed Engineering at AI-native B2B companies.
How it works
Three steps from install to a deputy that knows every engagement on your roster.
Connect Slack, Linear, GitHub
Three OAuth installs. Outpost reads your customer-channel history, Linear projects, and GitHub eval-failure checks.
Spawn a deputy per engagement
Each customer-engagement gets its own deputy with durable per-engagement memory across weeks of context.
Approve drafts in Slack
The deputy drafts status updates and triage notes — your FDE reviews and ships. Nothing leaves your workspace without approval.
Two workflows. Done by Monday.
Outpost ships with the two highest-leverage drafts an FDE writes every week. Approve in seconds. Audit-logged end-to-end.
Friday Customer Status Draft
- Posts every Friday 3pm local in the customer Slack channel
- Pulls the week's progress, blockers, and next-week plan
- Self-reports confidence so the FDE knows what to double-check
Weekly status — Acme × LaunchWings
Shipped: prompt v3, eval coverage 78→84%
Blocked-on-customer: PII redaction sign-off (Legal)
Next week: cutover plan + load test
confidence 0.82 · n=5
Eval-Failure Triage
- Watches eval regressions from Linear tickets & GitHub PR checks
- Proposes cause, fix, and a new eval case to pin the regression
- Comments the triage note for FDE review — never auto-merges
Eval regression: refund-intent precision 0.91 → 0.78
Probable cause: prompt v3 dropped the few-shot for partial-refund variants.
Proposed fix: restore 2 examples in refund.system.md.
New eval case: “partial refund w/ store credit” (added to evals/refund.jsonl).
Pricing
Priced per customer-engagement. Pay for the work, not the seats.
Design Partner
First 5 design-partner customers. 60-day paid pilot, weekly product loop with the founder.
- Up to 1 engagement to start
- 60-day paid pilot
- Founder on Slack
- Co-shape the v1 roadmap
- Migrate to List pricing after pilot
List
Annual commit. For teams past the pilot stage who want a deputy on every active engagement.
- Minimum 3 engagements
- Annual commitment
- Slack, Linear, GitHub integrations
- Audit-logged deputy actions
- Priority support
Per customer-engagement, not per seat. Engagement = one of your customers your FDE team is actively supporting.
Designed for the FDE who has to sign off
Outpost is a deputy, not an autonomous agent. The human stays in the loop on everything customer-facing.
Human-in-the-loop
The deputy drafts. Your FDE approves every external action. Nothing posts to your customer's Slack without explicit sign-off.
Per-engagement memory
Each deputy has durable memory scoped to one customer-engagement. Context persists across weeks; nothing bleeds across customers.
Audit-logged
Every deputy action — drafts, edits, approvals, posts — is persisted and exportable. Compliance and security teams can review the trail.
Frequently asked
If we missed yours, email akash@lyric.tech.
How is this different from a chatbot?
Outpost is a persistent deputy spawned per customer-engagement, not a chat surface. It runs scheduled work (Friday status drafts), watches eval-failure signals, and posts drafts into your existing Slack / Linear / GitHub surfaces. The deputy carries context across weeks; a chatbot starts over every conversation.
Does Outpost ever post to my customer without approval?
No. The deputy drafts in your internal Slack and waits for an FDE to approve. Nothing leaves your workspace to a customer surface without an explicit human sign-off. Every approval is audit-logged.
What data does the deputy ingest?
For v1: the customer's Slack channel history you grant access to, Linear projects/tickets scoped to the engagement, and GitHub PR-check signals for eval failures. No code execution on customer data. No CRM / Notion / Confluence / Gong ingestion in v1.
Can we self-host?
Not in v1. The deputy runs on our infrastructure with per-engagement isolation. Self-hosted / VPC deployment is on the roadmap for design partners who need it; ask us in the pilot conversation.
When can we start a pilot?
Design-partner slots are open now. Pilots are 60 days at $1,000 / engagement / month, with a weekly product loop with the founder. Book a slot via the button at the top of this page.
What if we already use Slack's built-in AI?
Slack's AI summarizes channels. Outpost owns recurring work — drafting the Friday update, triaging an eval regression, keeping per-engagement memory across weeks. Different job. Many teams run both.