AI
An agent with your workspace in reach
Ask Runa. It reads your workspace to answer.
Runa is the assistant inside Subnomic. Give it an agent access and it can query a database, read a pod’s logs, check a rollout or publish a site — each step shown with the arguments it used, never with more permission than you have yourself.
- Tools
- 100+
- Models
- Anthropic · OpenAI
- Runs as
- a scoped access
- Free credits
- 2k / month
How it works
Decide what it may touch, then ask.
Create an agent access
Pick the scopes and the targets — a database, a cluster, a bucket. An expiry is required. The access can never hold more than the person who created it, and that is checked again on every call.
Ask in the chat
@-mention a database, server, cluster or a live Kubernetes object to pin it. Pick the model per message, and switch on extended thinking for Anthropic models.
Read what it did
Every tool call is a step with its name, arguments, output, status and duration. Changes go through the same task pipeline as a person’s, and land in the same logs.
What Runa can do
More than a chat box with a search bar.
A hundred-odd tools
Databases, Kubernetes, Docker, hosts, sites, storage, incidents, alerts, schedules and workflows — offered only as far as the access allows.
Investigations, not just lookups
Crash-loop reports, rollout history, recent changes near an incident, and security reviews of RBAC, secrets and exposed workloads.
Pinned context
Mention what the question is about and Runa starts from it. The picker only offers what the access can reach, and the server checks again.
Guardrails apply
A guardrail set to deny or require approval refuses the call — Runa cannot talk its way past a rule a person wrote.
Your own AI clients over MCP
Connect Claude Desktop or Cursor to the workspace with an agent token. Scopes, targets and expiry work the same way as in the chat.
Bring your own key
Use the platform model, or give the workspace its own Anthropic or OpenAI key — or an OpenAI-compatible endpoint.
Incident investigation
When an incident opens, Runa looks at what changed and what is failing, and writes the postmortem when it is resolved.
Summaries and plain-language search
Recorded terminal and database sessions summarised, and the activity log searchable with a question instead of a filter.
Proposed fixes wait for a person
From an investigation Runa can propose restarting or scaling a workload. Nothing runs until somebody with the right to do it approves.
What it runs as
An access you can read in one glance.
Runa never borrows your session. It acts through an agent access: a named credential with scopes, targets and an expiry. What it may do is the access’s scopes intersected with your own permissions, so a demoted author quietly takes it down with them.
- Without an access, the chat can search the activity log and nothing else.
- An access that cannot restart a deployment means Runa cannot either.
- Servers that require approval to connect need an active just-in-time grant — Runa can request one, never approve it.
- Tool output and attached files are treated as data, not as instructions.
Limits
What each plan includes.
| Plan | Free | Pro | Team | Enterprise |
|---|---|---|---|---|
| AI credits / month | 2k | 100k | 500k | 1M |
| Agent accesses | — | 2 | 10 | 100 |
| MCP for external clients | — | — | ✓ | ✓ |
| Proposed fixes from incidents | — | — | ✓ | ✓ |
A thousand input tokens cost one credit and a thousand output tokens five; thinking counts as output. Unused credits roll over. The whole price list.
Questions
Before you give it an access.
Does it ask before it changes something?
It asks through the access you gave it. If the access cannot make a change, neither can Runa, and a guardrail that requires approval refuses the call. Fixes proposed from an incident always wait for a person.
What does the model provider receive?
Your question and the conversation so far, facts for anything you mentioned, attached files, and the results of the tools it called during the turn. Its own reasoning is kept for you to read and is not sent back.
What can it do on the free plan?
Search the workspace’s activity log in plain language. Tools need an agent access, and those start on Pro.
Can I choose the model?
Per message, from your provider’s model list. A workspace can also use its own Anthropic or OpenAI key instead of the platform default.
The rest of the workspace
Same bill, same credentials, same audit log.
Ask it something only your workspace knows.
Create a workspace, give Runa an access, and ask. The free plan includes 2,000 credits a month.