---
title: "What Is Model Routing in Tracer? Rate-Limit Fallback"
url: "https://goremy.ai/blog/articles/tracer-model-routing-rate-limit-fallback"
author: Luis Chavez-Mattos
editor: Luis Chavez-Mattos
editor_title: Director, Product
published: "2026-11-17T14:40:11.166Z"
updated: "2026-11-17T14:40:11.166Z"
topic: AI Pricing & Costs
description: "Tracer's model routing keeps an AI coding chat going when a provider hits a rate limit. Learn the fallback options, what is off by default, and the caveats."
---

# What Is Model Routing in Tracer? Rate-Limit Fallback

> Tracer's model routing keeps an AI coding chat going when a provider hits a rate limit. Learn the fallback options, what is off by default, and the caveats.

Edited by Luis Chavez-Mattos, Director, Product · Published 2026-11-17

## What is model routing in Tracer?

Model routing is a setting in Tracer, a free and open-source desktop app for running coding agents, that keeps a chat alive when your AI provider hits a usage limit or goes down. You build a fallback plan: try another account on the same provider, switch to an equivalent model on a different provider, wait for your limit to reset, or notify you if nothing works. It is off by default, so you have to turn it on.

## TL;DR

- Model routing lets a Tracer chat **keep going after a rate limit** by moving to a backup account, a similar model, or a wait-and-resume.
- You set the **order of the fallback steps** and decide which models are allowed to stand in for each other.
- **It is off by default.** Nothing changes until you switch automatic routing on in settings.
- It covers **chat agents on the selected host only**, not terminal agents.
- A switch **starts a new agent session seeded with your chat history**, so the new model has context but is not a continuation of the same session.
- The wait option only works if the provider **reports a reset time** that falls within the maximum wait you set.
- If every fallback fails, Tracer **notifies you** so you can step in.

## Why does a rate limit break an AI coding chat?

AI coding tools run on a provider's servers. Providers cap how much you can use in a given window, and sometimes their service simply goes down. When either happens, the agent stops answering, usually in the middle of a task.

If you work with one agent on one subscription, the fix is manual. You wait, or you open a different tool, paste in the context, and explain where you were. That costs time, and the explanation you write is never as complete as the original conversation.

The problem grows if you pay for more than one plan. Tracer is built around bringing your own subscriptions, so you can run agents such as Claude Code, Codex, OpenCode and Cursor side by side without buying another plan from Tracer. Having several subscriptions means you often have somewhere else to go when one runs dry. Model routing is the part that makes that switch happen without you doing it by hand.

## How does Tracer's fallback plan work?

Open settings and go to model routing. There you build a plan from a set of steps. Tracer tries them in the order you choose.

| Fallback step | What it does | Good for |
|---|---|---|
| Another account on the same provider | Moves the chat to a second login with the same provider | People with two subscriptions (say, a personal and a work plan) |
| Equivalent model on another provider | Switches to a model you have marked as a substitute | Provider outages, or when one vendor's limit is used up |
| Wait for the usage limit to reset | Pauses and resumes when the limit clears | Short limits where switching models would change results |
| Notify you | Alerts you if no other step worked | The last line of defense |

Two controls sit on top of the steps.

**Which models can replace one another.** You decide what "equivalent" means. A model you trust for careful refactoring might be a fine stand-in for another strong model, but you may not want it replaced by a cheaper, weaker one mid-task. The mapping is yours to set.

**Whether the chat switches back.** You choose whether the chat returns to the original provider once it is available again, or stays on the fallback.

If a switch is pending and you don't want it, you can cancel it.

## How do you turn on model routing?

1. Open Tracer's settings.
2. Go to **model routing**.
3. Turn on automatic routing. It is off by default.
4. Build your fallback plan by ordering the steps you want.
5. Choose which models may replace one another.
6. If you use the wait option, set the maximum wait you will accept.
7. Decide whether chats should switch back to the original provider when it recovers.

A sensible first pass is a short plan: second account first, equivalent model second, notify last. Add waiting only if you are comfortable pausing.

## What are the catches?

The details matter here, because they decide whether routing helps you or surprises you.

**It is off by default.** If you install Tracer and never visit the setting, a rate limit will stop your chat as it always has.

**It applies to chat agents, not terminal agents.** Routing works on chat agents on the host you select. Agents you run in a terminal are not covered, so a terminal session that hits a limit still needs you.

**A switch starts a new session.** When Tracer moves you to another account or model, it starts a new agent session that uses the chat history. The new model gets the context, but it is a fresh session reading the history, not the same session carrying on. Expect it to pick up well in most cases, and expect to check its first steps on a complicated task.

**Waiting depends on the provider.** The wait option only works when the provider gives a reset time, and only when that time is within the maximum wait you set. If the provider gives no reset time, or the reset is further away than you are willing to wait, that step is skipped and the plan moves on.

**"Equivalent" is a judgment call.** Different models behave differently. A different provider's model may format code differently, make different assumptions, or handle your project's quirks less well. Tracer lets you choose substitutes, but it cannot promise identical results.

## Which fallback order makes sense?

The right order depends on what you value. These are suggestions, not Tracer defaults.

- **You have two accounts with the same provider:** put the second account first. It is the least disruptive switch because the model stays the same.
- **You have subscriptions with different providers:** put the equivalent-model step next, and be selective about which models you allow.
- **Your limits reset within a short window:** consider placing the wait step before a cross-provider switch if you care more about consistency than speed.
- **You are away from your desk:** keep the notify step at the end so you hear about a stalled chat rather than discovering it later.

If you are doing delicate work, such as a large migration, you might leave routing off for that task, or allow only the same-provider account and waiting. For routine work like writing tests or tidying documentation, a wider set of substitutes costs you little.

## What does this mean for you?

**If you code with AI agents daily,** this removes a common interruption. The most likely benefit is for people who already pay for more than one plan and have been treating the second one as a manual escape hatch.

**If you manage a team that uses AI coding tools,** it is a reason to ask how people handle limits today. Do they share accounts, hold personal subscriptions, or just stop? A fallback plan is easier to reason about when it is written down.

**If you are not a developer,** the lesson still applies to any AI tool you rely on at work. Usage limits and outages are normal. Know what your backup is before you need it.

What to do:

1. List the AI subscriptions you actually have access to.
2. Decide which models you would accept as stand-ins for each other.
3. Turn on routing and test it on a low-stakes task first.
4. Review the first few switches to see how well the new session picked up the context.

## Could you build a small tool to track your AI limits?

Fallback plans work better when you know which accounts you have and how often they run out. A team juggling several AI subscriptions could describe a simple internal tracker to [Remy](https://goremy.ai), which builds and deploys full-stack business apps from a plain-English description. It might hold each account, who owns it, which models it covers, and a log of when limits were hit. That would replace a shared spreadsheet that nobody updates, and the team would own the result instead of paying for another subscription.

## Frequently Asked Questions

### Is Tracer's model routing on by default?

No. Automatic routing is off by default, so you need to open settings, go to model routing, and turn it on. Until you do, a rate limit will stop your chat as usual.

### Does model routing work for terminal agents?

No. It applies to chat agents on the host you select, not to terminal agents. If you run an agent in a terminal and it hits a limit, you will need to handle it yourself.

### Do I lose my context when Tracer switches models?

Not entirely. When it switches, Tracer starts a new agent session that uses the chat history, so the new model has the context. It is still a new session, so check its first actions on complex tasks.

### Can Tracer just wait for my limit to reset?

Yes, with conditions. The wait option only works when the provider gives a reset time and that time falls within the maximum wait you have set. Otherwise Tracer moves to the next step in your plan.

### Can I stop a switch I don't want?

Yes. If a switch is pending, you can cancel it. You can also choose whether a chat returns to the original provider once it is available again.
