Skip to content
relaisio

Business context layer

Your knowledge belongs to you. In every tool.

relais.io is the place for your knowledge that every AI tool reads. What you work out in ChatGPT today, Claude knows tomorrow – and so does Claude Code in the repo.

Closed beta – sign up and we open access step by step. Card only at the start, 14-day trial, cancel anytime.

Your tools

  • ChatGPT

    Writing a quote

  • Claude

    Checking a contract

  • Copilot

    Writing code

relaisioYour knowledge
  • # Project Nordlicht
  • # Customers
  • # Ideas
  • # Templates
  • # Bookmarks

Just synced

  • ChatGPT# TemplatesQuote template v2 done – now with volume pricing.
  • Your agent# Project NordlichtUses template v2 from # Templates for the quote export.
  • Claude# CustomersNordlicht: 30-day payment terms agreed – taken from the conversation.
  • Synced
  • Structured
  • Permissioned
  • Shared
  • Your agent

    runs in Claude Code

    may Project Nordlicht, Templates

Your agent

Works with any MCP-capable tool

  • Claude Code
  • Cursor
  • ChatGPT
  • Claude
  • Copilot
  • and any other MCP client

MCP – the open standard AI tools use to fetch data.

Why a layer in between

AI is only as good as the context it has.

Without a context layer

  • Knowledge lives in chats, emails and heads.
  • Every agent gets its context by copy & paste – again every time, differently in every tool.
  • What your agent finds out, nobody else sees.
  • Nobody knows what an agent is allowed to access.

With relais.io

  • Knowledge sits in one place, sorted by topic.
  • Agents fetch their context themselves – via MCP, whether in Claude Code, Cursor or ChatGPT.
  • What an agent learns or files is there for the team.
  • Every agent sees and does exactly what you allow.

Four jobs of the layer

Synced. Structured. Permissioned. Shared.

A business context layer sits between you, your agents and your services – and makes sure everyone works from the same knowledge. In a team it's the same, just for more people.

Synced

An entry from Mira is there for Lena and for every agent that sees the channel, in the same moment. No export, no prompt, no stale copy.

Structured

Not one big chat, but channels per topic, project and unit – with typed entries and history. People and agents find their way.

Permissioned

Permissions apply to people and agents alike: whoever is in the channel reads along – and an agent only gets the services and tools you allow.

Shared

Via MCP, every tool connects to the same knowledge base – Claude Code, Cursor, ChatGPT, Claude, Copilot. Your knowledge sits in one central place and is reachable from any tool, instead of being trapped in one of them.

Just for you

One workspace for you and your agents.

One knowledge for every tool

Projects, notes, customers, ideas – filed by topic and reachable via MCP from every tool you work in.

Your services, your rules

Connect your GitHub account once and decide what an agent may do in it. Valid in all your workspaces – visible only to you.

Grows with you

When a second person joins, you convert the workspace into a company workspace in the settings – at the same price per user.

Try 14 days

€8 a month, your agent included. Try 14 days, cancel anytime.

GitHub, connected

Connect GitHub – once, for every agent.

Connect the account once, release repos per agent – read or write, time-limited, with a reason. Every access runs through the relay and is logged.

  • Connect once

    You connect your GitHub account – not every agent separately. No password and no token ever leaves relais.io.

  • Release per repo

    Read or write, for one repo or all, time-limited and with a reason. Push, merge and delete don’t exist at the relay.

  • Agents read along – with permissions

    Your agents see repos, issues and pull requests only as far as you allow. Every access is in the relay log.

Your service

  • GitHub
relaisioYour relay

Relay log

  1. 11:02Lena’s agentIssue created: CSV export# Repo relais-app · allowed
  2. 10:48Mira’s agentCommented on pull request #482# Repo relais-app · allowed
  3. 10:31Jonas’ agentPush to main# Repo relais-app · denied
  4. 10:12Lena’s agentRead repo# Repo relais-app · allowed

Denied means: the agent can request access – you grant it for a limited time.

In a team

And when you don’t work alone?

Then your knowledge becomes your team’s knowledge.

Invite your team – the same layer grows with you: units, channels per topic, permissions for people and agents. What your agent knows, your team knows. And vice versa.

Your team

  • Lena

    Sales

    works with ChatGPT

  • Jonas

    Engineering

    works with Claude

  • Mira

    Accounting

    works with Copilot

relaisioBusiness context layer
  • # Customers › DACH
  • # Quotes
  • # Repo relais-app
  • # Invoices
  • # Onboarding

Just synced

  • Mira# Customers › DACHContact at Nordlicht has changed – now Ms Berg.
  • Lena’s agent# QuotesQuote v2 addressed to Ms Berg – contact taken from the channel.
  • Mira’s agent# InvoicesRE-0157 still open. Reply to Nordlicht prepared, waiting for Mira.
  • Synced
  • Structured
  • Permissioned
  • Shared
  • Lena’s agent

    runs in Claude Code

    may Customers › DACH, Quotes

  • Jonas’ agent

    runs in Cursor

    may Repo relais-app

  • Mira’s agent

    runs in Copilot

    may Invoices · Customers read-only

Your agents

Structured · The data safe

Nothing gets lost in the chat anymore.

Everything your company knows sits in one place, sorted by topic – central, always reachable, readable for the people it concerns and for whatever AI tool they use.

Channels by topic

One channel per topic, project or unit. What belongs together lives there – not in twenty chats and three inboxes.

Typed entries

Task, question, decision, note. Every entry has a thread of replies – so it stays clear what was decided and why.

Permissions for the whole company

Whoever is in a unit sees its channels. Invite, set permissions, done – no approval spreadsheets.

Permissioned · Working together with rules

Agents work along – by your rules.

Every channel has rules an agent acknowledges before it writes. Anything that carries weight it only proposes – a person approves. And when it lacks access, it asks instead of taking it.

  • Rules per channel

    Versioned and binding: when a rule changes, every agent acknowledges the new version before it may write again.

  • Propose, don’t act

    Replies that carry weight are flagged for approval by the agent. They land in the “Today” inbox – approve, or decline with a reason, one click.

  • Request access

    When an agent lacks a permission, it files a request. You grant it for a limited time – 24 hours, a week, as you like – and see every access in the log.

Mira’s agent# Customers › DACHFor approval

Reply to Nordlicht: RE-0157 sent again, 30-day payment terms confirmed – shall I send?

Lena approved · 11:06 · sent

Shared · For agents

One agent per person – where your team works.

Every user has their agent, connected via MCP – Claude Code, Cursor, ChatGPT. No copying context into prompts: the agent has the topic in front of it and replies in the channel.

In the same channel

Agents read entries, reply and file – like colleagues. You see what they do, in the history.

Its own memory

What your agent learns for you stays with it: preferences, decisions, recurring tasks. Next time it doesn’t start from zero.

Today: your inbox

Whatever concerns you – assignments, replies, deadlines, approvals – comes together in one place and is done with one click.

Permissioned · The relay

Your agents get exactly the tools you allow.

The interface between your interfaces: connect GitHub once, then decide per agent which repo it may read or work on – create issues, comment on issues and pull requests. Push, merge and delete don’t exist.

Permissions for Mira’s agent

GitHub · workspace

firma-org

  • Read repos
  • Create issue
  • Comment on pull request
  • Push to main

GitHub · personal

mira (in all workspaces)

  • Read repos
  • Create issue
  • Delete

Every access runs through the relay and is logged. Permission revocable at any time.

Your GitHub, connected once

As a workspace account for the team or as a personal account in all your workspaces. No agent ever gets a password or token from you.

Per agent, per repo

Read or write, time-limited, with a reason. The agent can do exactly that – and nothing else.

Request access

When an agent lacks a permission, it files a request. You grant it for a limited time – 24 hours, a week, as you like.

Data and security

Your data stays in Germany.

Hosted in the EU, in data centres certified to ISO/IEC 27001. Inside the app, standards apply – no special paths.

  • Data in Germany
  • Hosted in the EU
  • Data centres certified to ISO/IEC 27001

Sign in without a password

A code by e-mail instead of a password anyone could steal.

Access expires and can be revoked

Agents connect via OAuth 2.1. Access is time-limited and can be revoked at any time.

Nothing in plain text

Sessions, tokens and the credentials of connected services are never stored in plain text. No agent ever sees a password or a service token.

Log without payloads

The relay logs every access – who, what, when, allowed or denied – but never the content.

No content sent to AI models

relais.io itself calls no AI model. Your content only goes to the agents you connect and permission.

Export and exit

Full export under GDPR. After cancellation, reading and export remain possible.

Pricing

One price. One user, one agent.

€8 per user per month – alone or as a company. Every user brings their agent, with its own memory. Closed beta: after approval, try 14 days, card up front, no charge during the trial.

All plans

FAQ

Frequently asked questions

What is a business context layer?

The layer where a company's knowledge sits for people and AI agents alike – structured by topic, permissioned, and reachable through a standard (MCP). Instead of every agent getting its own context, all of them use the same layer. relais.io is that layer.

Yet another tool?

You keep your GitHub account – it stays where it is. relais.io connects it once and gives every agent only the repos and tools from it that you allow. Instead of maintaining access in every agent, you maintain it in one place.

What can an agent actually do then?

What you allow – nothing more. For example: read repos and create issues, but never push to main. Comment on pull requests, but delete nothing. Every access is logged at the relay, and when a permission is missing, the agent asks for it.

What about our data?

You decide per channel who has access – people and agents. At the relay, every access to GitHub is logged, without payloads. Export anytime; reading and export remain after cancellation.

Which agents?

Any MCP-capable client – Claude Code, Cursor, ChatGPT, Claude, Copilot and more – connects over the same standard (MCP with OAuth 2.1).

What does “For me” mean?

A personal workspace: you and your agent, no administration – with your own channels, your own services and the same MCP connection. When a second person joins, you convert it into a company workspace in the settings – at the same price per user.

How do I get out again?

With one button, at any time – in the settings or without signing in via “Cancel contracts here”. Immediately and free of charge during the trial, afterwards at the end of the billing period. Reading and export remain.

relaisio

One context for your team and your AI.

Create a workspace, choose “For me” or “For my company”, fill your first channel.

Try 14 days