Pi on 600+ models, through one extension
Install the Requesty extension for the Pi coding agent and every model your key can use appears in /model. Claude sessions are cached automatically, throttled requests fail over, and spend lands in one dashboard.
pi install git:github.com/requestyai/pi-requesty # ~/.pi/agent/auth.json { "requesty": { "type": "api_key", "key": "$REQUESTY_API_KEY" } } pi --provider requesty --model claude-sonnet-5-5
Set up Pi in three steps
The extension registers Requesty as a provider and loads the models your key can use every time Pi starts, so there is no model list to maintain.
- 01Install the extension
Run pi install git:github.com/requestyai/pi-requesty. Update it later with pi update --extensions.
- 02Add your key
Add a requesty entry to ~/.pi/agent/auth.json. The key can be an environment variable, a password manager command or the key itself.
- 03Pick a model
Start pi, run /model and type requesty to filter the list, or pass --provider requesty --model on the command line.
{ "requesty": { "type": "api_key", "key": "!security find-generic-password -s requesty -w" } }
Every model your key can use, in /model
The list follows your key's approved models: discrete models, your organization's routing policies and managed models, including @eu variants.
If you override baseUrl in models.json, keep the /v1 suffix: the extension loads the model list from <baseUrl>/models.
off the input price, at most, for tokens served from a provider cache. Requesty caches this agent by default.
of provider failures on the gateway in April 2026 were 429s. Throttling, not downtime, is what stops an agent mid-task.
eventual success on a routing policy in April 2026, against 85.01% for callers pinned to one provider.
Claude sessions cached without a flag
The extension identifies Pi to Requesty, which turns on automatic prompt caching. On Anthropic models Requesty adds the cache breakpoints, so Pi's system prompt and tool definitions are reused across turns, cutting cost by up to 90% on long conversations.
- Cache hits you can seeCached tokens show on each request after the first turn
- Fallback on 429sUse a policy as the model and a throttled request retries elsewhere
- A cap on the keyA runaway loop stops at a monthly limit you chose
Share of April 2026 requests where the upstream provider returned a failure. Throttling, not downtime, is what stops an agent mid-task. See the data note.
Pi with Requesty, answered
How do I add a custom provider to the Pi coding agent?
Install the Requesty extension with pi install git:github.com/requestyai/pi-requesty and add your key to ~/.pi/agent/auth.json. Pi then lists Requesty as a provider with every model your key can use, so you do not write a models list by hand.
How do I fix rate limits in the Pi coding agent?
Rate limit errors come from the upstream provider. Start Pi with a Requesty fallback policy as the model, for example pi --provider requesty --model policy/coding-default, and a throttled request moves to the next model in the chain. Requesty itself does not cap requests per minute.
Which models can Pi use through Requesty?
Everything your key is approved for: discrete models such as anthropic/claude-opus-5-5, managed models such as gpt-6.1-terra, @eu variants and your organization's routing policies. Run /model and type requesty to filter the list.
How do I see token usage and cost for Pi?
Every request appears in the Requesty analytics dashboard within seconds with tokens, cost and latency. After the first turn on an Anthropic model you should see cached tokens above zero, which confirms prompt caching is working.
Are there free models for the Pi coding agent?
Requesty includes 200 requests a day on free models with no credit card. Use the same key in Pi, and paid models are billed per token at the provider price plus 5% when you need them.
Can Pi read my Requesty key from a password manager?
Yes. In ~/.pi/agent/auth.json, a key value that starts with ! runs a command that prints the key, so any password manager CLI works and the key never sits in a file. Pi runs the command when it starts.
Can Pi use the EU endpoint?
Pick an @eu model variant, such as claude-sonnet-5-5@eu, from /model. To send all Pi traffic through the EU gateway, set baseUrl to https://router.eu.requesty.ai/v1 under providers.requesty in ~/.pi/agent/models.json.
Guides for the other agents you run
The same key and the same dashboard work across every harness, so spend and logs stay in one place.
Run Pi on Requesty
600+ models, a spend cap on every key, and every session on the record.