Model routing
In pilotRoute each step to a model or to plain code based on task type, cost and data sensitivity. Running in our own engineering operations.
Lamiak Core · Agents
Agents that propose, inside limits you set.
Bounded agents that draft and propose rather than act, with each step sent to a model or to plain code depending on what the step is.
The question
What is an agent allowed to do on its own?
Draft and propose. The agents built so far prepare one piece of work — a reply, a follow-up, a set of changes to a record a person then applies or rejects — and stop there, waiting for a person. Which model handles a step, or whether plain code handles it instead, is decided by routing, and routing is the one part of this running in our own engineering operations.
How it works today
In developmentWhat is built, stated as built.
Routing runs in our own engineering operations. The agents that draft sit inside the workspace and are in development.
A model, or plain code
Routing picks per step, on task type, cost and how sensitive the data is. Plenty of steps do not need a model at all.
One task at a time
An agent is handed a single task and the packet assembled for it, rather than the run of a workspace.
It stops at a draft
The drafting agents built so far queue their work for a person to approve or reject, and releasing an approved message is still being built.
What is planned
PlannedWhat is designed, stated as designed.
Agents that propose inside a customer's own systems are designed and not built.
Agents in your systems
Agents that propose actions in the tools a business already runs on, with a person approving each one and the decision recorded.
Past drafting
Ticket triage, access requests and runbook steps, each prepared by an agent and released by a person.
What this leans on
The registry entries behind this page.
Each card carries its own status, taken from the registry rather than written here. How the labels are used is set out on the trust page.
Approval-gated email drafts
In developmentReplies to new client inquiries are drafted from incoming email and held until a person reviews and approves the exact message; sending approved replies is still being built.
Human approval
Follow-up drafts
In developmentAn agent prepares client follow-up drafts and queues them for a person to approve or reject; sending approved follow-ups is not built yet.
Human approval
Voice-memo updates
In developmentA spoken update is turned into proposed changes to the client record, which wait for a person to apply or reject them; the review screen is not built yet.
Human approval
Approval-gated agents
PlannedAgents that propose actions inside your systems, with a person approving and a record of each decision.
Human approval
IT operations agents
PlannedTicket triage, access requests and runbook execution with approvals.
Human approval
The rest of Lamiak Core
The parts this one sits beside.
- ContextWhat the system knows before it acts.Assembling the records, documents and permissions a task needs into one packet, before an agent is allowed to do anything with them.
- Workflows and approvalsSteps that run the same way, and the person who decides.Ordinary repeatable steps, with a person approving where the policy for that workflow says so, and each decision recorded with who made it and when.
- Evaluation and learningMeasure first, then decide what changes.Evaluation suites and graders run by hand today. Scorecards, promotion gates, dataset snapshots and training from outcomes are designed and not built.
- Deployment and governanceWhere your records sit, and what governs a change.How records are kept apart per organisation, which outside services receive data and why, and what would have to be true before something changes.
Start with one workflow that matters.
We map the work, the systems it touches and the approvals it needs before anything is automated.