📝 Written ● Beginner Updated 2026-05-13

How do I switch providers mid-conversation in LingCode?

TL;DR: Click the provider dropdown in the chat header and pick a different model — LingCode carries the conversation history into the new provider. Use when one model stalls on a refactor: send the same chat to a different provider without losing context.

When Claude stalls on a refactor, hand the same chat to DeepSeek or GPT-4 without starting over. LingCode carries the conversation history into the new provider so you keep your context and lose only the agent's mood.

The most common reason an AI agent fails on a task isn't that the task is too hard — it's that the model you happen to have selected is the wrong one for the job. Some models excel at long, careful refactors and stall on quick text edits. Others are fast and cheap but lose the thread in complex multi-file changes. The honest answer to "which model should I use" is: it depends on the turn, and you'll know after the first response. Sticking with a model that's clearly struggling, just because you started the chat in it, is a habit worth breaking.

The expensive way to switch is to open a fresh chat in the other provider and re-explain everything you've already said. That works but it's tedious — and any nuance from earlier in the conversation is lost. The cheap way is to switch providers in place, mid-conversation, and let the new model inherit the entire history. The new provider reads what was said, what was tried, what was rejected, and picks up the next turn with full context.

LingCode runs three independent agent loops underneath — one for Claude, one for DeepSeek, one for thirteen OpenAI-compatible providers — and serializes the chat history into whatever format the next provider expects. From your seat in the chat panel, it's a dropdown click. What you'll see in this tutorial is when that click pays off, what carries over and what doesn't, and when starting fresh is the better call.

What you'll learn

Why "wrong model for the task" is the most fixable failure mode
How to switch providers from the chat panel
What carries over (history) and what doesn't (tool state)
When to start fresh vs. switch mid-flight

Step 1: Notice when a provider is wrong for the task

The four signals

Switch providers when you see any of these:

Repeats itself. The agent keeps proposing the same edit despite "no, do X instead." Often a model that's anchored on its first answer.
Loses the plot on long edits. Forgets earlier constraints in long multi-file edits. Different models have different effective context.
Slow on the model you have, faster elsewhere. If you're on a thinking model and the task is mechanical, a faster non-thinking model finishes in a third the time.
Tool-use error loops. Same tool, same wrong argument, three times. Some models reason their way out faster than others.

Step 2: Switch from the provider selector

From the chat panel

At the top of the chat panel there's a provider selector — the dropdown showing the current model (e.g. "Claude Opus 4.7" or "DeepSeek V4 Pro"). Click it and pick a different provider/model from the list.

The chat panel reloads with the new provider active. Your existing message history stays in the panel. Send your next turn and the new provider receives the full conversation as context.

Available providers: Claude (default), DeepSeek, LingModel (built-in, free), GPT-4 / OpenAI, Gemini, Mistral, Groq, plus 7+ other OpenAI-compatible providers and a BYO endpoint slot. Each needs an API key in Settings — except LingModel.

Step 3: Understand what carries over

History yes, in-flight tool state no

What carries to the new provider:

Every prior user/assistant message in the turn — verbatim.
System prompt / instructions for the workspace.
Files referenced in the conversation (the new provider re-reads as needed).

What does NOT carry:

An in-progress tool call. Switching mid-tool-call cancels the call cleanly — the new provider sees the canceled turn and decides whether to retry.
Provider-specific cache state. Anthropic prompt caching, OpenAI session memory, etc. — each provider starts cold.

Step 4: Sometimes start fresh instead

When the history is the problem

If the chat is full of failed attempts, the new provider inherits all that noise. Long failed-trail histories drag any model down. In that case, hit New chat, paste a clean summary of the actual goal, and start the new provider on a clean slate.

A rough rule of thumb: if you've sent more than ~10 corrections in a row, the conversation is poisoned. Switch providers AND clear history.

Step 5: Pair specific providers with specific tasks

What each provider tends to be best at

Claude — long, careful refactors; structured tool-use; agentic loops that span many turns.
DeepSeek — cost-efficient general coding; strong on Python and mainstream Swift; fast.
GPT-4 / GPT-5 — broad knowledge, good for unfamiliar frameworks; strong UI design judgment.
LingModel — built-in, zero setup, free; route routine tasks here to save your paid keys for hard problems.
Gemini — strong at long-context retrieval; useful for "find everywhere we use X" in big repos.

These are tendencies, not laws — every release rebalances. Form your own preferences by switching often.

Cost note: Switching providers also switches whose API meter is running. If you're escaping Claude into DeepSeek because of cost, great. If you're escaping LingModel (free) into GPT-4 (paid) just to retry something, your wallet now matters — check the small print of the provider you switch into.