$ cat ~/map/pi-hyper-provider.md
pi-hyper-provider
Pi extension that registers Charm's Hyper inference provider with API key or OAuth, dynamic model discovery, and optional Hypercredit balance in the status line.
extensions✓ Oct 16
Facts
- GitHub Stars
- ★ 29
- Forks
- 6
- license
- MIT
- npm / mo
- —
- last push
- Sep 24, 2026
install
pi install npm:@charmland/pi-hyper-provider## TL;DR
Pi extension from Charm that registers the Hyper inference provider — API key or OAuth login, dynamic model discovery persisted to pi's model store, and an optional Hypercredit balance and team name in the status line.
## Key points
- $Registers Charm's Hyper inference provider with pi; published on npm as @charmland/pi-hyper-provider and installable from git as well.
- $Two auth paths: OAuth via pi's /login (pick Subscription, then the Charm Hyper provider), or set the HYPER_API_KEY environment variable before opening pi.
- $Dynamic model discovery: after credentials are configured, pi discovers available Hyper models and persists the latest successful catalog in its standard model store; a previously discovered catalog remains available when a later refresh fails.
- $Pick models with /model, filtering by provider name hyper or model names like glm-5.1.
- $/hyper-status configures the status line interactively (teamName defaults to false, hypercredits to true) or directly with /hyper-status teamName true, /hyper-status hypercredits false, and /hyper-status reset.
- $Caveat for headless users: as of pi v83, provider extension model caches are only created and updated when pi is launched interactively — open the TUI to pull new models or refreshed metadata.
## FAQ
Q How do I authenticate with Hyper?
For OAuth, open pi, send /login, pick Subscription, and select the Charm Hyper provider. For an API key, set HYPER_API_KEY in the environment, then open pi.
Q Why are new Hyper models not appearing?
Since pi v83, model caches for provider extensions are only created or updated on interactive launches. If you mostly run pi headless, open the TUI whenever you want new models or updated metadata.
## References
## Related entries
$ big-pi
Model-driven compression that summarizes completed phases into tiered blocks while Pi's JSONL transcript stays authoritative and untouched.
$ Pi Taste
Local preference-learning extension that distills accepted and rejected turns into one readable taste.md.
$ pi-agent-substrate
Suspendable remote workspaces for pi and Oh My Pi, on a self-hosted Agent Substrate cluster.
## Guides
Last updated: Aug 31, 2026