Policies in plain language
Write what an agent must never do, read or say, with examples of where the line is. Policies can check what a person sends, each action an agent takes, and its reply.
Governance
Your rules around every model call and every action.
Agents act on real systems. RocketKind puts your team’s rules in the path of everything they do: what they may say, what they may touch, what needs a person, and what they may spend.
No promises outside policy. Never promise a refund, credit or delivery date that our policy doesn’t allow.
Crosses the line“I’ll refund you in full today, no questions asked.”
Fine“I’ve asked a lead to approve your refund.”
Judged by the model you choose, including TypeSafe AI’s Jev
One model for all of it
One security review, not six.
Agents, workflows, data, APIs and the model gateway share one identity, approval, audit and spend model. SOC 2 is in progress.
Who can do what
- Single sign-on, roles and permissions
- Agents act as a service account or as the person asking
- People grant every connection
- Secrets handed over by link, never through chat
- Encryption under your keys
Before it goes live
- Validation on every save
- Test runs before release
- See what uses it before you change it
- Publish a version, roll back in one step
While it runs
- Approvals by kind of action
- Spend limits and budgets
- Audit trail and change history
- A trace of every run and model call
Guardrails
Guardrails your team writes, checked where you choose.
Our governance and architecture brief covers how checks, approvals and budgets behave in detail. Talk to us for a copy.
Judged by the model you choose
The agent’s own model, a model from any provider you connect, or TypeSafe AI’s Jev, a judge built to answer with probabilities instead of prose.
Approvals for what matters
Choose which actions an agent takes on its own, which ask a person first, and which it never takes.
Limits on every request
Cap how far an agent can go on one request, so nothing runs away with your time or your money.
Budgets
Spend caps for every agent, workspace and key.
Caps for each agent
Set what an agent may spend in one conversation and in one day.
Everything counts
Spend from the tools, workflows and agents an agent uses counts against its cap.
Ceilings above agents
Set ceilings for each workspace and the whole organization, and token and spend limits for each API key.
Spend you can see
Every model call is priced and attributed to the agent, workspace and key that made it.
Security
Built for the systems you can’t afford to get wrong.
Permissions on every asset
Roles, teams and per-asset sharing, with Admin, Manager, Editor and Viewer roles in every workspace. Nobody can grant more than they have.
Encryption with your keys
Secrets and sensitive fields are encrypted at rest with AES-256-GCM, under platform keys or your own in AWS KMS, Google Cloud KMS, Azure Key Vault or HashiCorp Vault, with audited rotation.
Private Network Tunnels
An agent you run inside your network opens an outbound connection, so RocketKind reaches internal systems without inbound ports or a VPN.
Run it where you need it
Managed by us on the cloud you choose, deployed into your own cloud account, or on-premises.