All use cases

Use case · Software companies and product teams

Your documentation, finally within reach of your IDE agent.

Ts-Knowledge syncs your GitHub repositories — documentation first, code if you ask for it — extracts services, APIs and dependencies, and serves it all to your developers' agents over MCP. Every answer cites the file and the section.

The docs exist. Nobody can find them.

  • Twelve repositories, three wikisThe production deployment procedure is in there somewhere — in a README, a documentation page, a closed ticket.
  • “Why is there a message queue here?”The decision is two years old. Its author has left; the reasoning sits in an architecture document nobody ever opens.
  • The IDE agent guessesIt knows the code that's open, not the other services' docs. So it makes up the name of an endpoint that doesn't exist.

1 · Find

The exact passage, cited down to the file and section

A repository resyncs without rereading everything: only the files that changed go through again. Documentation and code aren't split the same way: the section heading gives each excerpt its context.

A procedureHow do I deploy the billing service to staging?
The relevant section of the deployment guide, with the repository, path and heading that locate it.orbel-billing / docs/deploy.md / Staging
An identifierWhere is INVOICE_LOCKED defined?
Found by its exact name — exactly where a meaning-only search would miss it.orbel-billing / docs/invoices.md / Invoice states
“I don't know”What is the rate limit on the public API?
Not enough evidence: no page documents it. The agent says so instead of inventing a number.

2 · Understand

Your services, their APIs and dependencies, drawn from the docs

With the “Software” template, Ts-Knowledge reads your documentation and extracts services, APIs, endpoints, databases, message queues, teams — and what connects them: calls, depends on, stores in, exposes, maintained by, replaces. Each one comes with the passage that proves it.

Services and components

Their language, framework, repository, version and status — all sourced.

Dependencies

Who calls whom, who publishes to which queue, who stores data in which database: a graph you can navigate.

Owners

Which team maintains which service, and since when — without copying an org chart.

Decisions and migrations

Releases, deprecations, migrations, decisions: the platform's events in order, each with its source.

The Knowledge Snowflake arranges a project into facets based on a profile: there is one for fiction series and one for e-commerce, but not yet one for a software platform. The graph and the timeline, however, are already there.

3 · Connect

In the IDE, the agent searches before it writes

I'm changing how credit notes are calculated. Who consumes the invoice events?

get_entity_relationsbilling-eventsconsumed by: notifications, accounting · published by: billing
search_knowledge“credit note event format” · Orbel Platformorbel-billing / docs/events.md / Credit notes

Two services consume the queue: notifications and accounting. The format of the “credit note” event is documented in docs/events.md · Credit notes — keep its fields, or give both teams a heads-up.

  • One key per agent, restricted to the spaces you choose, with search and read access by default: an IDE agent doesn't modify the corpus.
  • The same index as the REST API: your internal tools query exactly what the agent sees.
  • Structured citations: repository, path, section, branch URL — everything you need to reopen the file.

Get started in four steps

  1. Create the space

    One space per platform or per team, depending on who should see what.

  2. Connect GitHub

    Authorize access in one click, then pick the repository and branch. Documentation by default, code with a single checkbox.

  3. Turn on knowledge

    The “Software” template: services, APIs and dependencies build up with every sync.

  4. Connect the IDE

    One key, the MCP URL in the agent's configuration — and it searches before it writes.

  • IDE agents
  • Claude
  • ChatGPT
  • Gemini
  • Mistral
  • Your own code, via the REST API

Have your agents read the docs before they write the code

Connect a repository, let the sync run, and ask the question a new hire would ask.