Docs · Guide
Reasonix Studio on any model
Reasonix Studio 2.x and its CLI share one config.toml and .env. magpie adds a provider named magpie there, pointing at magpie's local gateway, so Reasonix's executor and planner can each be any model you have in magpie: DeepSeek, GLM, Kimi, Qwen, GPT or a ChatGPT subscription. Your other providers, keys and project or session settings are left alone.
Set up Reasonix
- Open magpie's Agents pageFind the Reasonix Studio row. Its model field is called executor.
- Pick a magpie modelThe list has two groups: Reasonix Studio's own models and the ones via magpie. Pick one of the latter; magpie writes the provider and key, then sets
default_modeltomagpie/<model>. - Set effort and a planner if you wantA model with effort levels can take an effort. The planner can be a different model, or off to turn the separate planner off.
magpie reasonix magpie/deepseek/deepseek-v4-pro # executor magpie reasonix effort high magpie reasonix planner magpie/deepseek/deepseek-v4-flash # a separate planner, or: off
What magpie writes
Reasonix's folder is REASONIX_HOME; without it, %APPDATA%\reasonix on Windows and ~/.reasonix elsewhere. A Reasonix in WSL uses Linux's ~/.reasonix, with the gateway as the distro reaches it. magpie writes:
# ~/.reasonix/config.toml default_model = "magpie/deepseek/deepseek-v4-pro" [[providers]] name = "magpie" kind = "openai" base_url = "http://127.0.0.1:3425/v1" api_key_env = "MAGPIE_REASONIX_API_KEY" models = ["…", "…"] default = "deepseek/deepseek-v4-pro" reasoning_protocol = "openai" no_proxy = true [providers.model_overrides."deepseek/deepseek-v4-pro"] supported_efforts = ["…"] context_window = … max_output_tokens = … # ~/.reasonix/.env MAGPIE_REASONIX_API_KEY=…
Each model carries its context window, max output, image support and effort levels; a model with no effort levels gets reasoning_protocol = "none". When magpie's models change, magpie updates the provider. When Studio has a [desktop] provider_access list, magpie adds magpie to it. The key goes only in .env; config.toml names the variable.
magpie only touches its own entry. If there is already a magpie provider magpie didn't make, another provider uses MAGPIE_REASONIX_API_KEY, or .env sets it twice, magpie stops with an error rather than overwrite. magpie keeps a restore record in its own config folder (settings only, never an upstream credential) to put your default_model back when you switch back.
Executor and planner
The executor is default_model and the planner is [agent] planner_model; each can come from magpie or from one of Reasonix's own providers. While the planner still uses magpie, magpie's provider stays, even with the executor back on your own model.
When it takes effect
Reasonix Studio 2.x and the native 1.39.x/2.x CLI read global settings at startup, so start a new process after a change. A model set for a project or session overrides the executor and planner.
Switching back
Switching back undoes only what magpie added: the magpie provider goes, default_model returns to what it was before (or goes if there was none), magpie leaves provider_access, and magpie's key leaves .env; a file magpie created is removed once empty. A model you picked later, and edits to other providers and keys, stay. You can also just pick a model from the Reasonix Studio group.
magpie reasonix default
FAQ
How do I use DeepSeek, GLM or a ChatGPT subscription in Reasonix Studio?
On magpie's Agents page, pick a via-magpie model as Reasonix Studio's executor. magpie adds a magpie provider to config.toml, the key to .env, and sets default_model to magpie/
Can the planner be a different model?
Yes. The planner field writes [agent] planner_model: another magpie model, one of Reasonix's own, or off.
I changed it and Reasonix didn't notice. Why?
Reasonix reads global settings at startup. Start a new Studio or CLI process; a model set for a project or session also overrides the global one.
Will my own providers and keys change?
No. magpie adds and removes only the provider named magpie and MAGPIE_REASONIX_API_KEY, and stops with an error on a same-named entry it didn't make.
More agents on the same models: jcode on any model, Crush on any model and Models in OpenCode. Something fails? See Fixing common errors, or ask on Discord.