AI knowledge layer · for B2B teams

Every answer your company already knows. One ask away.

Edgespan connects your databases, wikis, and tools into a single AI analyst. Plug in Postgres, ClickHouse, Notion, GitHub, Slack — get an answer with sources, in seconds.

Book a demo → Self-host →
edgespan / live
›
thinking — querying postgres, searching notion, ranking results
The problem

Your company knows the answer. Just not where.

Every business has the same pattern: data in one warehouse, knowledge in another tool, and the only people who can connect them are too busy to ask.

01 / Bottleneck

Analysts are the chokepoint

Every "quick question" from leadership becomes a 3-day SQL ticket. Decisions wait, urgency dies, the dashboard never gets built.

02 / Sprawl

Knowledge lives in 12 places

Notion, Confluence, GitHub, Slack threads, three spreadsheets nobody opens. The truth exists. Finding it does not.

03 / Onboarding

New hires re-learn for months

The institutional context that makes your senior people effective is locked in their heads — and in DMs from 2022.

How it works

Three steps. No data team required.

Edgespan is a single agent built on a graph of specialized models. It picks the right tool for each question — SQL, retrieval, or both — and returns a cited answer.

STEP 01

Connect

Plug in your sources via OAuth or read-only credentials. Postgres, ClickHouse, BigQuery, Snowflake, Notion, Confluence, GitHub, Slack, Drive — and growing.

STEP 02

Define

Describe your metrics and entities once in a semantic layer. What "active user" means. Which joins are canonical. Edgespan never hallucinates a definition again.

STEP 03

Ask

Anyone in the company can ask anything in plain language — Russian, English, mixed. Every answer comes with the SQL it ran and the documents it cited.

Your sources
postgres / prod
clickhouse / events
notion / company-wiki
github / docs
slack / #product
edgespan
Your answer
› why did MRR drop in March?
MRR fell 8.2% MoM driven by the Pro tier — 47 churned accounts, 38 cited the new pricing. Confirmed in #product retro thread (Mar 28).
Integrations

Plug in everything. Or just what you use.

Native connectors maintained in-house. Read-only by default. Permission inheritance from each source — if a user can't see it there, the agent can't either.

postgres
mysql
clickhouse
bigquery
snowflake
redshift
notion
confluence
github
gitlab
slack
telegram NEW
gdrive
s3
metabase
superset
looker SOON
dbt SOON
Use cases

Built for everyone who waits on data.

Edgespan replaces the Slack message that starts with "sorry to bother you, can you pull..."

For CEOs and founders

  • What's our cash runway at the current burn?
  • Show MRR by tier for the last 6 months
  • Which accounts are at risk this quarter?
  • Summarize what the team shipped last week from GitHub and Linear

For product managers

  • What's the activation rate for users who came from the iOS launch cohort?
  • Which features have the lowest engagement among Pro users?
  • Find every Notion doc and Slack thread about the onboarding redesign
  • Compare retention of users with and without integrations enabled

For growth and marketing

  • What's the LTV by acquisition channel for Q1?
  • Which campaigns drove the highest-quality signups last month?
  • Show me the funnel from landing visit to paid for the EU geo
  • What's our CAC trend across paid channels?

For customer support

  • Find the docs page about SSO setup for SCIM users
  • What did engineering say about the API rate-limit issue last week?
  • Show all open tickets mentioning "billing" with priority high
  • What's the resolution time SLA for enterprise customers?

For new hires

  • Who owns the payments service and what's the on-call rotation?
  • Walk me through how we deploy to production
  • What's our sales playbook for mid-market deals?
  • Show me every doc and discussion about our pricing strategy
Security · self-hosted

Your data never leaves your perimeter.

Edgespan ships as a Docker Compose deployment you run on your own infrastructure. Source code reviewed. No telemetry. No third-party data sharing — ever.

Built for regulated industries and security-conscious teams. AIFC, fintech, healthcare, government — Edgespan is the only AI knowledge platform that runs entirely inside your VPC by default.

Cloud also available for teams that prefer managed. SOC 2 Type II in progress. ISO 27001 mapping included.

⌬

Read-only by default

Every database connector uses scoped read replicas. Writes require explicit approval flow.

⊞

Permission inheritance

If a user doesn't have access to a Notion page or table, the agent never sees it for them.

◊

Full audit log

Who asked what, which SQL ran, which documents were retrieved. Exportable for compliance.

▣

BYO model

Anthropic, OpenAI, Bedrock, Azure, or self-hosted Llama / Qwen. Your keys, your traffic.

Pricing

Simple. Predictable.

Per-seat, with a fair-use cap on tokens. Self-hosted is flat-rate. No bait pricing, no usage cliffs.

Starter
$29/ user / mo
For small teams getting started with their first 2–3 sources.
  • Up to 25 users
  • 3 connectors
  • 50k queries / month
  • Email support
  • Cloud-hosted
Start free trial
RECOMMENDED
Growth
$79/ user / mo
For teams running on data — full source coverage, semantic layer, eval suite.
  • Up to 200 users
  • Unlimited connectors
  • Unlimited queries
  • Semantic layer + custom metrics
  • SSO, audit log, priority support
  • Cloud or self-hosted
Book a demo
Enterprise
Custom
Self-hosted, custom connectors, dedicated success engineer.
  • Unlimited users
  • Custom connectors built for you
  • BYO model + private inference
  • SOC 2, ISO mapping, MSA
  • SLA + dedicated Slack channel
  • VPC, on-prem, air-gapped
Talk to sales
Questions

The honest answers.

Glean and Onyx index documents — they answer questions from wikis and chat. They don't query your database. Tools like WrenAI and Vanna do text-to-SQL but ignore your knowledge sources. Edgespan is built from day one to combine both: it knows when to write SQL, when to retrieve documents, and when to do both in the same answer.
Yes. All database connectors are read-only by default and connect through a scoped role. Every query is statically analyzed before execution — SELECT-only, mandatory limits, cost estimation against BigQuery's dry-run API. You can also point Edgespan at a read replica or a snapshot warehouse for full isolation.
The semantic layer is the answer. You define your metrics, entities, and canonical joins once in YAML — the agent is constrained to those definitions. Every answer ships with the SQL it ran and the documents it cited, so you can verify in one click. Internal evals run on a golden query set with each release.
Never. Self-hosted deployments don't send anything outside your network. Cloud deployments use BYO model keys (Anthropic, OpenAI, Bedrock) which all support zero-retention modes — we configure them on by default.
First connector takes 10 minutes. A useful semantic layer for your top 5 metrics takes a couple of hours. Most teams are getting real answers from real users within a week. Enterprise self-hosted deployments typically go live in 2–3 weeks including security review.
Edgespan answers in English, Russian, German, Spanish, French, and a dozen more — including mixed queries. The semantic layer is language-agnostic; users can ask in any language and get answers in the same one.

The edge that spans
your entire company.

Book a 20-minute demo. We'll connect to one of your sources live and you'll see the answer to a real question by the end of the call.

Book a demo → Read the docs