Solution · Access
Which models each seat may use
Which models each seat’s agent may call, with what data, through one gateway.
Model ask, then Daniel decides, then Gateway.
Who decides · Daniel Ruiz · CTO · a new model for a seat
How it runs
Every model call from every seat goes through one gateway. Which models a seat may use, and whether customer data may reach them, is a rule like any other: owned, versioned, and asked for through a request.
A seat is a role a person and an agent hold together; the escalation matrix says who decides above each limit, seat by seat.
SEC-R09
customer data only to approved models
- Person decides11:00Sofia asks for a faster model for the service agent’s review packets
- Agent carries11:05Request drafted: the model, the data it would see, the cost against the Payments team’s limit
- Rules check11:06
SEC-R09
customer data only to approved models
- Person decides15:00Daniel approves for code and test data only; the gateway applies it
Your issue tracker · Platform
REQ-2290
A faster model for review packets
- Assigned to
- SBInvoicing service seat
- Status
- Needs Daniel
Illustration · Halden, a fictional company
- 01Sofia Brandt, service ownerAsks for a faster model for the service agent’s review packets.
- 02Her agentDrafts the request: the model, the data it would see, the cost against the team’s limit.
- 03Priya Nair, security leadReviews it, because customer data is a question (SEC-R09).
- 04Daniel Ruiz, CTOApproves it for code and test data only.
- 05The gatewayApplies the decision; every call from the seat goes through it.
Where the decisions sit
Decision, who decides, up to where, and above that whom.
- A new model for a seatDecides: the CTOUp to: the approved modelsAbove that: the CEO
- Customer data to a modelDecides: the security lead reviewsUp to: approved models only (SEC-R09)Above that: the CTO
- Model spend above a team’s limitDecides: the department headUp to: the team’s limitAbove that: the CTO
How much the agent decides, to start. Asks every time: a new model for a seat is a decision.
The record it leaves. The request, the review, the decision, and every call from the seat since.
How it plugs in
No new inbox, no new tool to learn.
- 01
An app in your tools
Connects to your repositories, trackers and chat as an app
- 02
A check, not a meeting
Holds a decision as a check the owner clears
- 03
Where people already work
Asks the owner where they already work, and carries out what was decided
HAModel access
Which models each seat’s agent may use
Coding agents on the models Halden chose
Changes go through a request
The models Halden approved
Customer data only to approved models
The models Halden approved
Answers reviewed before they leave
Connected
The tools it runs through
- Coding agents
Agents you builtClaude CodeCursorGemini CLIGitHub CopilotOpenAI Codex
- Model providers
AnthropicAmazon BedrockAzure OpenAIGoogle GeminiOpenAIModels you host
We connect to the systems you run, and build the connection before the demo.
Start with howyou run today.
Bring one piece of work that breaks at the handoffs: a pull request waiting on three reviewers, an initiative spread across twelve tools, a release nobody owns. We model the seats it crosses, give one of them an agent, and follow one decision to its record.