Intlo Labs · engine builds
Institutional memory that fills itself.
Your company already knows the answer. It’s in a Slack thread from May, an email chain with the client’s counsel, a pull request nobody outside the team read. The person who knows is in a meeting, on holiday, or gone.
Figures from our own production deployment, not projections.
What it is
A knowledge base that reads the systems your team already uses, distills every thread into institutional memory, and answers questions with citations that link back to the source. Nobody files anything. It never guesses.
How it works
- 01
Ingest
Live workers and scheduled syncs read your systems. Threads settle before capture, so the conclusion is stored rather than the opening message. Edits update; deletions remove. Code is chunked on function boundaries, and only what changed is reprocessed.
- 02
Distill
Each thread is condensed to question, summary, resolution and references. Low-signal material is filtered before it reaches the index — on our own two years of email, roughly 93% was correctly set aside as noise.
- 03
Retrieve
Meaning-based and keyword search run together, weighted for recency and fused by rank, with the question automatically rewritten and routed to the sources most likely to hold the answer.
- 04
Answer
A cited answer, with a hard rule: when the material does not contain the answer, it says so and shows the closest related evidence instead.
Connects to
Slack, email and GitHub are live today. Every other system is a single connector module: CRM, helpdesk, Notion, Drive, your data warehouse. New source in, same memory out.
It reaches you as an MCP server — so Claude consults company memory mid-task — a password-protected web app, and a REST API for your own tooling.
Built to be trusted
Cited or silent
No answer without evidence, and no confident guessing.
Measured, not asserted
Before go-live we build fifteen real questions with your team and report how often the correct source lands in the top three — including the misses.
Auditable
Every filtering decision is logged, so you can inspect what was set aside rather than take it on faith.
Resumable
Kill any sync mid-run; cursors resume without duplicates or gaps.
Observable
A status dashboard shows every source's freshness, row counts and model spend.
Extensible
Each source is a single connector module — CRM, helpdesk, docs, warehouse. New system, same memory.
What we need from you
Read access to the systems in scope, the bot invited to the channels you want read, fifteen real questions your team actually asks, and two decisions: how far back to reach, and what stays out.
Data and governance
Your content stays in a database you own. We document exactly what is ingested, which processors see it in transit, and how to have anything removed from the index. Where email is in scope, we walk your team through what that means for third-party correspondents before a single message is read.
$2,499 to build. $99 a month to run.
Fixed price for a standard three-system deployment, no per-seat pricing, and model usage billed at cost on your own status page.