Pricing
Fixed to build. Flat to run.
One price to stand it up on your systems, one price to keep it running. Model usage is billed at cost and shown on your own status page, so there is never a surprise line item.
- ·Discovery, scoping and access design
- ·Build across three systems
- ·Full backfill of your existing history
- ·Accuracy evaluation on your own questions
- ·Branded web app and assistant integration
- ·Runbook and team training
- ·Hosting and live workers
- ·Monitoring and security updates
- ·Support and channel changes
- ·Re-running your accuracy set as the corpus grows
- ·No per-seat pricing — your whole team uses it
A standard deployment covers three systems. Additional connectors are quoted per system. Model and embedding usage is billed at cost.
What the build looks like
Eight stages, each ending in something you watch work. Nothing proceeds until the previous stage passes its check.
- 0
Access & scope
You create the bot, authorise read-only access, and we agree what stays out. About 90 minutes of your time.
- 1
Foundation
Your database, source registry and resumable ingest ledger. Gate: migrations run clean and repeatably.
- 2
First source & search
Usually documents or code — the largest corpus, so retrieval work surfaces early. Gate: you search your own material and get real results.
- 3
Live chat ingestion
The worker connects and backfills approved history. Gate: you post a thread and watch it become searchable, then edit it and watch the entry change.
- 4
Email
Incremental sync and the noise filter. Gate: we review a sample of the filter’s decisions together before the full backfill runs.
- 5
Answers, assistant & API
Routing, retrieval and cited synthesis. Gate: cross-system answers with citations, and an honest refusal to an unanswerable question.
- 6
The app your team uses
Branded web app and status page. Gate: your team signs in on a phone and a laptop.
- 7
Accuracy & handover
We measure your fifteen questions, tune, and report the score including misses. Gate: the number is on paper and your team can operate it.
Start with a 30-minute call
Enough to scope it: which systems, how far back, and the questions your team keeps re-asking.