Models

Magic Coder routes every conversation through a model. Which model — and which provider — is configurable per session and per message.

Picking a model

There are three ways to pick a model:

• At launch — coder --model <name> sets the model for the new thread.
• In the TUI — /model opens an autocomplete picker with everything your BridgeApp account has access to.
• As a default — set [generation] model = "<name>" in ~/.magic/config.toml.

Specifying a model at launch overrides the user default. Picking with /model overrides for the current thread until you switch again.

The /model picker

/model

Opens a fuzzy-search list of available models. Each row shows the display name, internal name, and provider, e.g.:

GPT-5.5 (gpt-5.5) — openai (default)
Claude Sonnet 4.6 (claude-sonnet-4.6) — anthropic
Grok 4.20 (grok-4.20) — xai

When two rows share a model name, the provider in the trailing field tells them apart.

Type to filter: /model claude narrows to Claude variants, /model grok matches by model name, /model anthropic matches by provider. Fuzzy match is supported — /model snnt will surface claude-sonnet-4.6.

To go back to the user default:

/model default

Setting a user default

Edit ~/.magic/config.toml:

[generation]
model = "gpt-5.5"

After that, every new thread uses gpt-5.5 unless you override at launch (--model …) or in the session (/model …).

You can also clear the default by deleting the line, or by selecting /model default once and then editing the config out — Magic Coder writes through /model choices to the user config.

Available models

The list of models you can pick from comes from BridgeApp — it’s the set your account is entitled to. Common providers include:

• OpenAI
• Anthropic
• xAI

Specific model names (sizes, generations, reasoning vs. non-reasoning) change as new releases land. The /model picker is always the source of truth for what’s available to your account today.

If you /model and the picker shows “No models found,” your BridgeApp account doesn’t have any models active. Talk to your BridgeApp admin or check your account at bridgeapp.ai.

If you /model and you’re not signed in, the picker shows “No models available yet (/login first).” Selecting it will start the login flow.

Switching mid-thread

You can switch models in the middle of a thread. The new model picks up the conversation context and continues. This is useful for a few patterns:

• Cheap drafting, expensive review. Start with a fast small model, then switch to a stronger one before producing the final patch.
• Different strengths. Switch from one provider’s model to another’s mid-task to cross-check a tricky decision.
• Cost discipline. Drop down to a smaller model for routine grunt work, switch up for the hard part.

The conversation history is the same; only future messages run through the new model.

Model and reasoning effort

Magic Coder treats model and reasoning effort as separate dials. Picking a model doesn’t reset reasoning effort, and changing reasoning effort doesn’t change models. See Reasoning effort.

Some models map reasoning levels differently — for example, Anthropic keeps xhigh native on Claude Opus 4.7 but degrades it to max or high on older supported models. Magic Coder handles the translation automatically; you just pick the level that matches the work.

What “default” means

When you see (default) next to a model in the picker, that’s the BridgeApp-side default for your account — the one the engine uses if you don’t pick anything. Your user default (from ~/.magic/config.toml) takes precedence over the BridgeApp default. CLI flags and /model choices take precedence over both.