Private AI workspace

Ask your company.
See the evidence.

Onirix turns private company documents into cited answers and interactive, source-backed charts, using the AI models and infrastructure your organization chooses.

Remote work policyClaude

What is our international remote-work policy, and which exceptions need approval?

Employees may work from another country for up to 30 days per calendar year without approval1. Beyond 30 days, or in any country where the company has no payroll entity, the request needs sign-off from the manager and People Operations12.

Roles that handle regulated customer data need approval for any stay, regardless of length2.

The 2024 handbook still says 60 days3. The 2026 policy supersedes it, so the answer above follows the newer document.

Ask a follow-up
5
model providers, your credentials
9
document formats indexed
3
visibility levels, enforced in the query
0
outbound requests with Ollama

The product

Answers your team can check.

Every answer and every number stays connected to the file it came from.

See the story in your data

Numbers in spreadsheets and reports become charts. Each series cites its document, and the table is one click away.

  • Bar, line, area, pie, and scatter
  • Citations on the series
  • Charts are data, never generated code
Quarterly attainment by account executivebar
  1. 92%
  2. 83%
  3. 75%
  4. 67%
  5. 50%
  1. Okafor
  2. Lindqvist
  3. Moreau
  4. Tanaka
  5. Reyes

Attainment1Q2 pipeline.xlsx · sheet “AE summary”

TableCopy data

Verify every answer

Each claim links to the passage behind it, opened beside the answer, with a link to the original file.

Citations, passage panel, original link

Permissions apply before retrieval

Organization, team, or private visibility is enforced in the query. An admin does not silently read every team's documents.

PostgreSQL query and OpenSearch filter agree

Keep control

Self-host in Docker, bring your own provider, or run Ollama and send nothing outside your network.

Docker, your provider, or fully local

How it works

From a folder of files to a cited answer.

  1. 1

    Connect a provider

    OpenAI, Anthropic, Google, xAI, or Ollama. Enable several, set a default.

  2. 2

    Upload documents

    Originals are kept. Text is extracted, chunked, embedded, and indexed in the background.

  3. 3

    Ask a question

    Hybrid retrieval hands the strongest passages to the model. Answers stream.

  4. 4

    Check the sources

    Markers open the passage, its source, and its date. Missing evidence is said out loud.

Model providers

You pick the model.

Onirix uses your credentials, encrypted at rest. Nothing goes to a provider you did not choose.

  • OpenAI
  • Anthropic
  • Google
  • xAI
  • Ollama
  • Several at once, one default

Documents

Upload what you already have.

Up to 50 MB per file, with page and sheet anchors so a citation lands in the right place.

  • PDF
  • DOCX
  • XLSX
  • XLS
  • CSV
  • Markdown
  • HTML
  • JSON
  • Text

Where it fits

The questions teams ask.

  • What is our international remote-work policy, and which exceptions need approval?
  • Show quarterly attainment by account executive and mark the 85% target.
  • How does authentication work, and where are organization permissions enforced?
  • Compare the renewal terms in these agreements and call out conflicts.

Self-hosting

Your infrastructure. Your model.

The whole stack runs in Docker on servers you operate. With Ollama, nothing leaves the network.

docker compose6 services
  • webNext.js application
  • workerBackground indexing
  • postgresMetadata, chats, citations
  • opensearchHybrid vector and keyword index
  • redisIndexing queue
  • minioUploaded originals, or any S3 store
$pnpm run docker:up

Get started

Turn company knowledge into answers your team can trust.

Create a workspace, connect a provider, upload a folder.