Portage, by Souverain Tech

Knowledge is a journey.

Le savoir est un voyage.

A research environment for your organization's own field. What you learn goes into your organization's knowledge graph, and the graph grows as your research does.

  • Bring in documents, data feeds and web research. Each capture keeps its source, and goes in only once a person approves it.
  • Ask questions across the graph, and see the source behind every answer. Flag what it cannot answer for research.
  • Build datasets from the graph for your machine learning pipelines, and bring their findings back for review.
  • Decide who sees what (Public, Private or Counsel-only) on every question asked.

What it does

Carry it once. Find it every time after.

Bring it in

Each capture keeps its source, and goes in only once a person approves it.

Documents

Upload a document and a language model drafts what it says, statement by statement, each pointing to the lines it rests on. You review the draft, edit or cut any statement, and accept it. Nothing enters the graph until a person does. A draft can only speak about things your graph already knows, and email addresses and telephone numbers are left out.

Web research

When the graph cannot answer a question, flag it. Research reads the open web, working from the question and the names of the subjects your graph already knows, and cites every page it used: the address, the day it was read, and the words it relied on. Its findings come back as a draft for you to review, like a document's.

Data feeds and exports

News and job feeds (RSS and Atom), connectors of your own, and CSV exports fill datasets you can analyse, and the pages they link to can be read too, a few seconds apart. They feed your analysis, not the graph: what you decide matters is recorded as a capture by a person.

Ask, and see the source

Every answer shows where it came from.

A conversation with your graph

Ask in plain language. Each answer cites the captures it came from, and says so plainly when the graph does not know, so you can flag the question for research.

Saved questions

The questions your organization asks again and again are saved and reviewed once, and run for each reader at their own access level.

Your knowledge in your own AI tools

Connect your own AI assistants over the Model Context Protocol (MCP). They always read as the signed-in person, at that person's access level, and an outside tool never reads Counsel-only material unless you allow it.

Machine learning, under review

Model output never enters your graph on its own.

Datasets from saved questions

A dataset is a saved question's answers at a chosen access level, so it holds nothing a reader at that level could not see. Each one records when it was taken and a fingerprint of its rows.

Runs you can trace

Your models run against those datasets, and every run is tracked in MLflow with the data it saw. Results are shown beside the run, only to people whose access level allows.

People decide what stands

Your verdicts on a model's results become labels it learns from. A finding you accept is drafted as a capture for review, citing the run it came from. Nothing a model produces is recorded as fact until a person accepts it.

Who sees what

Public, Private or Counsel-only, on every question asked.

Enforced by the database

Every capture carries one access level. Each question is answered by the database from what the reader may see and nothing else, so a reader cannot even learn that something above their level exists.

Which AI models may see what

You decide which model providers may read Counsel-only material, if any. Material of that level is never sent to a provider you have not allowed.

Counsel-only, kept apart

Counsel-only material stays off the shared service unless you choose otherwise. If you do, it is kept in a database of its own, apart from everything else, and every read of it is recorded for your administrators.

Your organization's own addresses

Every Portage customer works under addresses of their own: their own name on souveraintech.com, or their organization's own domain if it has one.

your-name.souveraintech.com
portage.
Your front door: a public page, and your tools once you sign in
kb.
Your knowledge graph and its conversation
mcp.
Your knowledge, for your own AI tools

Plans

Personal

One person's knowledge graph, with its conversation and research.

Dedicated

An instance of your own: its own cluster, sign-in, vault and backups, sharing nothing with anyone.

On every plan, each customer's knowledge graph runs in its own database, and its files are encrypted under its own keys.

Pricing is set with each early-access customer. Ask us.

Start your knowledge graph

Early access is open to organizations and people in Canada. We set up your addresses and your sign-in, and show you around.

Request early access

Write to us at