Asgard AI

Sindri | Agent Execution

Where staff use agents every day

Sindri

Asgard Agents Hub

Use

Division of work with Odin

OdinAsgard Studio
IT sets an agent's prompt, the semantic models it can query, and the MCP servers and skillsets it can call in Odin, then publishes it to Sindri. Sindri has no screens for building agents.
ProjectIn Sindri
Each project has its own published agents. Switching projects switches the agents you can use.
Conversation workspaceSandbox
Agents query, compute and read or write files in a sandbox behind the conversation. The Files panel lets you browse, upload and download those files.
Approval
When an agent is about to write back to a system, an approval dialog shows the tool it wants to call and the exact input. Whoever holds the permission picks one: allow for this conversation, allow once, or deny. On a denial, nothing in the system changes.
Routines
Work finished in Sindri can be saved as a routine that runs on a schedule. When a run fails, Sindri notifies whoever owns it.
Wrong answers become test questionsContinuous improvement
An answer a user marks wrong becomes a rule and a new question in the test set. After a skill change the whole set runs again, and the change ships only if the total score improves.
Delegation@ an agent
Type @ to pull a named agent into the conversation. When you name no agent, Sindri routes the question to a suitable agent based on each agent's description.
Sub-agentsSubagents
When a job is too large, Sindri splits it across sub-agents and collects the results. Every step the agent takes is one click away.

Computer Use

Agents in internal systems

  1. 01

    One message

    A materials manufacturer built 14 internal web systems of its own. The platform runs 9 system agents, one per system. One message in Sindri hands off nine tasks: leave, work reporting, pallet orders, meeting rooms and more.
  2. 02

    The person's permissions

    The agent opens a browser to the portal and asks the person to sign in. Then the agent works with that person's permissions. It cannot see anything the person could not.
  3. 03

    API or browser

    The nine tasks go to nine system agents at once. APIs where they exist, the browser where they do not.
  4. 04

    Approval before writing

    Missing information is asked for with an option card. Anything that writes back stops for approval.

Product

Sindri in action

1 / 3

Arrow keys or swipe

A Sindri conversation that reports the result of nine tasks in a table
A conversation: one message asked for nine tasks. Sindri reports each result in a table and says which task still needs your decision. The screen comes from Sindri at a materials manufacturer and is de-identified.
The Sindri home page with the list of available agents
The home page lists the agents available in the current project. Each card says which kind of request it handles. The screen comes from Sindri at a materials manufacturer and is de-identified.
Recording: one message lists nine tasks. Sindri hands them to the agents for each system, asks when it needs a decision, and reports the result of each task. The recording comes from Sindri at a materials manufacturer and is de-identified.

Start here

Book a two-hour workshop

AI feasibility review of a live workflow