Use intent routing so a small model reads each turn and picks the model that fits it.
Advanced
1 hour
Claude Code
Codex
Seen in the wild
Evidence (similar to what people reported): X, Oct 2026
WhoDeveloper whose big model gets burned on simple questions like "what does this do?"
The problemEvery turn goes to the same expensive model, even trivial ones, which wastes quota.
The outcomeSimple questions go to a fast model while test-writing and heavier work go to the strong model.
Steps
Create a routing group containing a fast model and a strong model
Turn on intent routing for the group using the setup in the intent docs at usemagpie.ai/docs/intent
Point your agent at the group and ask one quick question, then one bigger task
Open the Routing view to see which model handled each turn
Prompt for your agent
Set up yetone/magpie intent routing so <FAST_MODEL> handles simple questions and <STRONG_MODEL> handles test writing. Confirm in the Routing view that two different turns went to different models.
Caveat. Intent routing needs a small model of your choice to classify each turn, which adds a call per turn. Exact setup steps are in the linked docs, not the truncated README.
WhoDeveloper who likes Codex or Claude Code but wants to try a cheaper model for everyday edits
The problemEach agent stores its model, keys and base URL in its own file and format, so trying another model means editing configs by hand.
The outcomeOne of your installed agents answers on the model you picked, chosen from a searchable list, with its config rewritten safely.
Steps
Download the app from usemagpie.ai, or run: curl -fsSL https://usemagpie.ai/install.sh | sh
Run: magpie provider add deepseek sk-<your-key>
Run: magpie codex moonshot/kimi-k2.5 (or: magpie claude deepseek/deepseek-v4-pro)
Start a new agent session and ask it to explain a file in your repo
Prompt for your agent
Install yetone/magpie, add a DeepSeek provider with the key in <DEEPSEEK_KEY_ENV_VAR>, then point Claude Code at deepseek/deepseek-v4-pro. Show me the command output, and verify by starting a new session that it is now on the new model.
Caveat. Changes apply to new agent sessions, not one already running. Preset names like deepseek are used in the README examples; confirm the exact model IDs in your picker.
WhoDeveloper with two Claude or ChatGPT accounts who keeps getting cut off mid-task by a 5-hour limit
The problemWhen one subscription hits its limit, the agent stops with an error and you lose your flow.
The outcomeThe agent keeps answering because requests move to the next account or model in your routing group, with no error shown.
Steps
Add each account or provider under Providers in the app
Run: magpie group add "Opus anywhere" models=claude/claude-opus-5-5,copilot/claude-opus-5.5 routing=smart
Run: magpie claude group/opus-anywhere
Run: magpie quota to see what is left on every plan
Prompt for your agent
Using yetone/magpie, create a routing group from my <PROVIDER_A> and <PROVIDER_B> models with smart routing and point Claude Code at it. Then run magpie quota and confirm both accounts show up as group members.
Caveat. Failover only helps if you really have more than one account or model available. Using several accounts may conflict with a vendor's terms, so check them first.
Check the Routing view to see the request arrive and which provider answered
Prompt for your agent
Using yetone/magpie, configure <TOOL_NAME> to use the local gateway at 127.0.0.1:3425 via its base URL environment variables. Run one request and verify it shows up in magpie's Routing view.
Caveat. A tool that hard-codes its API host or rejects custom base URLs will not work. One user also reported that some apps do not accept magpie as a provider.
WhoDeveloper paying for a ChatGPT or Claude plan who wants those models in a second agent without buying API credits
The problemA subscription works in only one agent, so paying for it twice or switching tools is the only way to reuse it elsewhere.
The outcomeA second agent runs on the models from your existing subscription, listed as a normal provider.
Steps
Sign in to your existing agent so magpie can find the subscription
Open the app and go to Providers, then Add provider, and pick the subscription
Open the Agents page and pick a model from that subscription for another agent
Start a new session in that agent and run a small task to confirm it answers
Prompt for your agent
Use yetone/magpie to expose my signed-in <SUBSCRIPTION_NAME> plan as a provider and assign one of its models to <TARGET_AGENT>. Verify the agent answers a short test prompt on that model.
Caveat. magpie reads local sign-in files and the Keychain to find subscriptions. Turn that off in Settings > Privacy > Subscription accounts, or run: magpie accounts discovery off.
Save a Budget or Focus profile and flip all your agents between them in one move.
Beginner
15 minutes
From the README
Any agent
WhoDeveloper who runs several agents and changes models by mood, deadline or budget
The problemChanging models across three agents means repeating the same edits three times.
The outcomeOne command swaps every agent between saved setups, such as a cheap Budget profile and a strong Focus profile.
Steps
Set each agent to the models you want for your budget setup
Run: magpie save work
Change the models, then run: magpie save focus
Run: magpie use work to switch every agent back at once
Prompt for your agent
With yetone/magpie, save my current agent setup as a profile named <PROFILE_NAME>, then change one agent's model and save a second profile. Show that magpie use switches back and the agent config reflects it.
Caveat. Profiles cover model choices for agents magpie manages. A running agent session keeps its model until you start a new one.
Import providers from CC Switch, Claude Code, Codex or Alma instead of re-entering every key.
Beginner
30 minutes
From the README
Any agent
WhoDeveloper already juggling provider configs across several tools who wants to consolidate
The problemRe-typing keys, base URLs and model lists into a new tool is slow and easy to get wrong.
The outcomeYour existing providers appear in magpie ready to assign to any agent.
Steps
Open the app and go to Providers
Choose the import option and pick the source tool
Review the imported providers and their model lists
Assign one imported provider's model to an agent on the Agents page
Prompt for your agent
Use yetone/magpie to import my providers from <SOURCE_TOOL>, then list which ones were brought in. Verify at least one appears in the model picker for an installed agent.
Caveat. Import covers only the four named sources. Imported keys are stored by magpie, so review them before relying on them.
Install an OpenCode or pi plugin so a sign-in flow becomes a usable provider for your agents.
Advanced
45 minutes
From the README
Any agent
WhoAdvanced user who wants Gemini or Antigravity models that need a plugin-based login
The problemSome providers only offer sign-in through agent plugins, which other agents cannot use directly.
The outcomeThe plugin's models show up in magpie and any agent can run on them like a normal provider.
Steps
Run: magpie plugin add opencode-gemini-auth
Run: magpie plugin login google-plugin
Open Agents and pick a model from the plugin's provider
Start a new session and send a short prompt to confirm it responds
Prompt for your agent
Using yetone/magpie, install the plugin <PLUGIN_NAME> and complete its login. Verify its models appear in the model picker and answer a test prompt in <TARGET_AGENT>.
Caveat. Plugins run on Bun, which is downloaded on first use, and plugin code runs on your machine. Only install plugins you trust.
Who it is forDevelopers who run more than one coding agent, provider or account and are tired of editing separate config files and hitting quota walls mid-task.
Why it is hotabout +651 stars a day since launch, 17k total, launched 26 days ago
DEDUPED