Manifesto

Your company already knows the answer.It just can’t find it.

Enclave is building the knowledge layer for the enterprise: one place to search and ask across everything a company knows, with every answer grounded in its sources, bound by existing permissions, and running where the data already lives.

The problem

Knowledge is everywhere.Answers are nowhere.

Every company runs on documents, threads, tickets, code and records spread across dozens of tools. The answer to most questions already exists somewhere. Finding it means knowing where to look, who to ask, and which version is current.

Search tools match keywords but miss how things connect. AI assistants answer fluently but rarely show their sources, and often need a copy of your data in someone else’s cloud. Neither is something a company can rely on.

What we believe

Enterprise search should be something you can trust.

01

An answer without a source is a guess.

Every answer should point back to the document, message or record it came from, and say how sure it is.

02

Permissions travel with the data.

Search should never show anyone something they couldn’t already open. Access rules come from the source systems, not from a second copy.

03

Connections matter as much as documents.

Knowing what a document says is half the job. The other half is knowing who owns what, what depends on what, and what was true on a given date.

04

Your knowledge stays in your environment.

Storage, keys, indexes and audit logs stay in your cloud account. That shouldn’t be a premium tier; it should be how enterprise AI is built.

05

Use the model you already trust.

Models change every quarter. The knowledge layer underneath them shouldn’t have to.

06

Measure in public.

We publish benchmarks, including the bugs we found and the claims we had to retire. Trust is earned with numbers, not adjectives.

What we are building

Four layers.One knowledge layer.

Scribe reads every document. The retrieval engine knows what each one says. Ontos knows how everything connects. Enclave turns all of it into answers people can check. Each layer stands on its own; together they make search a company can rely on.

  1. 01

    Enclave

    The answer engine

    One place to ask what the company knows. Cited, confidence-scored answers across Google Drive, Slack and GitHub, filtered by each person’s existing access.

    Building v1.0
  2. 02

    Retrieval engine

    What documents say

    A Rust shard store that keeps the index in your own object storage. Memory is only a cache, so the index never leaves your account.

    Benchmarked
  3. 03

    Ontos

    How things connect

    A knowledge graph of typed entities and relationships. Every fact keeps its source and its history, and every query leaves an audit record.

    Open source · Platform in alpha
  4. 04

    Enclave Scribe

    Reading every document

    A self-hosted document-understanding model that turns scans, PDFs, decks, handwriting and multilingual text into structured text, without sending pages to a third-party API.

    In training

What we are shipping

One honest releaseat a time.

  1. ShippedMay 2026

    Retrieval engine benchmark, v0.0.1

    Tested on a one-million-chunk corpus and real S3, with the cache bug we caught and the permission claim we retired.

    Read the report
  2. ShippedOpen source

    Ontos, the knowledge-graph layer

    Released under Apache-2.0 and installable with pip install enclave-ontos.

    View on GitHub
  3. Private alphaNow

    The hosted knowledge-graph platform

    Ontos as a managed service: create a project, upload documents, and query the graph over REST or MCP.

    Join the waitlist
  4. BuildingNow

    Enclave v1.0

    Drive, Slack and GitHub connectors, deployment into your AWS account, an employee app for asking, and an admin console for connectors, audit and tuning.

  5. In trainingNow

    Enclave Scribe

    Following the path Ontos took: built for Enclave first, then released as its own product once it clears the benchmark targets we’ve set for it. We’ll publish the numbers either way.

  6. NextAfter v1.0

    More sources, more clouds

    Confluence, Jira, Notion, Salesforce, SharePoint and Teams; GCP and Azure; and fully local models.

What we are looking for

Build it with us.

Design partners

Teams whose knowledge is spread across many tools and who want answers they can check. You get early access and a direct line to the people building it.

Talk to us

Early platform users

Developers who want a knowledge graph for their apps or agents without running the infrastructure.

Join the waitlist

Contributors

People who care about provenance, permissions and auditability in AI, and want to help shape the open-source layers.

Contribute to Ontos

Every company deserves answers it can trust.

Talk to usBack to overview