Keera Agents
Sandboxed AI agents that assist your team
Keera Agents runs AI agents for your business. Each agent is an assistant with its own sandbox, workspace and memory. Your team gives it work in Microsoft Teams, Slack or Google Chat, and every model call passes through Keera Gateway.
How Keera Agents works
Each task takes the same path, whoever starts it. Between tasks the agent sleeps, so an idle agent holds no compute.
-
Someone asks
In Teams, Slack, Google Chat or Keera Chat, on the board, on a schedule or with a new file.
-
The agent wakes
In its own sandbox, with its memory and files from last time.
-
It works
With its tools. Every model call goes through Keera Gateway.
-
It answers
Where the work came from. Then it sleeps until the next task.
What Keera Agents does
Each agent has instructions, a model, tools and a monthly budget.
Work from your chats
Write to an agent in Teams, Slack, Google Chat or Keera Chat. It answers in the same thread and remembers what was said there.
A board and a schedule
Assign it a card, and it moves the card to Review when it is done. Or let it run every weekday at eight and post the result to a channel.
Your folders
It reads and writes in the Google Drive, OneDrive and SharePoint folders you connect, and starts when a new file lands.
Your repositories
In GitHub or GitLab it pushes a branch of its own and opens a pull request for a person to review. Your main branch stays yours.
Agents under your rules
An agent is not a loophole. It follows the same rules as the person who owns it, and it only reaches what you connect.
Its own key
Every model call goes through Keera Gateway with the agent's own key, in its owner's name. Model access, budgets and the audit log apply as they do to people.
A sandbox of its own
Keera Agents runs on your existing Kubernetes cluster. Each agent gets a sandboxed container or a micro VM of its own, and what it runs stays inside.
No secrets inside
The agent's sandbox holds no password for your chats, drives or repositories. Keera Agents makes those calls and checks first that the agent may.
You decide who asks
Only its owner, everyone in the organisation, or the people you list. Anyone else gets a short no. When the owner leaves, the agent stops.
Questions about Keera Agents
What teams ask us before they hand work to an agent.
What does Keera Agents keep of my data?
What the work needs, and no more. Keera Gateway never stores the prompt or the answer, and nothing is trained on them; it keeps the request - who, when, which model, how many tokens, which decision - in your audit log. The agent keeps its memory and files in its own workspace, and Keera Agents keeps a log of each task for 90 days, so a person can check what the agent did. Deleting an agent deletes its workspace.
What can an agent in Keera Agents reach?
Only what you connect. It calls models and MCP servers through Keera Gateway with its own key, posts only in the chats it is connected to, reads and writes only in its folders, and pushes only branches that start with keera/. Its sandbox holds no password or token for those services: Keera Agents makes those calls itself, after checking that the agent may.
How does Keera Agents fit with Keera Gateway?
Keera Agents is built on the gateway. People sign in with their gateway account, and each agent calls models with a key of its own, in its owner's name. So model access, budgets, filters and the audit log apply to agents as they do to people, and each task shows in the gateway as one session.
Questions, comments, quotes
Have a question or a comment? Need a quote for on-premise or an SLA? Write to us.