Specialised Router
The right model for every step.
From your first instruction, the Specialised Router plans the work, runs each step on the right model through the tools you already use, and keeps it moving when one of them hits its limit. It is an optional part of Anuvrit.
From first instruction to finished work.
Plan, run on the right model, check and step up, hand off, finish to your stop line, and see what it cost.
Plan from the first instruction
In the tool you already use, the router works out the kind of work, asks Anuvrit memory what your team already knows, and returns a plan with the model and reasoning effort for each step and a one-line reason.
Run on the right model
Each step runs on the chosen model through your own installed tools: Claude Code, Codex, Cursor or local models. Claude Code can run a step on an OpenAI model through Codex, and Codex can run one on a smaller Claude model. One task can use several models, switching fast.
Check and step up
Each step is checked with your repository's own tests, lint and type checks. A failed check steps up one tier and reruns, and a review runs when the risk calls for it. Every outcome is recorded.
Hand off without re-explaining
The router keeps the task's state: goal, plan, progress, decisions and next step. When a tool hits its limit, the work continues, and you resume in any other tool with nothing to re-explain.
Finish to your stop line
When a session ends, the router finishes on its own, up to the stop line you pick for each repository: a diff, a commit or a pull request. It never pushes to a protected or default branch.
See the cost
Every task ends with a cost line, and a call log shows every counted call.
When a tool hits its limit, the work keeps going.
Smart handoff keeps the task, not the conversation, so the next tool starts where the last one stopped.
The state, not the transcript
The handoff packet holds the goal, the plan, progress, decisions and the next step. It is bounded in size, scanned for secrets and never contains transcripts.
Resume in any tool
Pick the task up in Claude Code, Codex or Cursor. The new tool reads the packet, so there is nothing to explain again.
Autonomy you control
Autonomy can be turned off or delayed for each repository. Before the router carries on by itself, you are notified with a chance to cancel.
Your agents, your machine. An optional add-on.
The router does not write code itself. It runs your own agents on the right model, and Anuvrit's cloud never holds your credentials or your code.
Credentials stay on your machine
The router uses the CLIs you have already signed into, or an API key if you prefer. Either way, the credentials stay on your machine.
Learns from outcomes
Outcomes are kept in Anuvrit memory, and the router draws on them when it plans the next task.
An optional add-on
The Specialised Router is an add-on to any Anuvrit plan. Anuvrit memory works the same without it.
A free taste every month
Every account gets a free taste of the router each month, with or without the add-on. The pricing page explains how router calls are counted.
Diagram: a closed loop of four steps. Work starts with company context, the Specialised Router assigns models, the outcome is checked, and the reviewed lesson is kept so the next task starts with more context.
Start with context. End with a lesson kept.
Anuvrit memory supplies the context, and the Specialised Router runs each step of the work on a model and reasoning effort matched to its difficulty and risk. The outcome is checked, and the reviewed lesson returns as knowledge the next task can use.
- 01Start with contextDecisions, reusable work and past lessons.
- 02Specialised RouterModels for planning, building and review.
- 03Check the outcomeTest, review and correct the work.
- 04Keep the lessonReviewed guidance, with evidence and scope.
A useful lesson should outlast the conversation.
The work draws on company memory. What you learn returns to it.
Lessons change behaviour through better context, not by retraining models. When we change how context is ranked, the change is measured on held-out cases first, released to a small share of requests and rolled back if it does not help. The Specialised Router learns from the same outcomes through Anuvrit memory.
Give your agents your team’s history.
Anuvrit is in beta and opening to engineering teams in stages. Join the waitlist and we will contact you when there is a place for your team.