Community Apps

Routstrd Auth

By Routstr

Self-hosted Routstr node with bearer-token auth

0.1.32 512MB Updated 14 Oct 2026
Package Website
ai proxy nostr bitcoin openai
Install this app
In your Cloudron dashboard, open the App Store, choose "Community apps" and paste this URL:
https://raw.githubusercontent.com/Routstr/routstrd-auth/refs/heads/cloudron/CloudronVersions.json

Description

Auth proxy and daemon for the Routstr AI routing network

Recent Changes

  • Auto-refill now checks the NWC balance before creating an invoice, and a refill invoice that times out is waited on until it reaches a final state before paying again, so a slow Lightning payment can no longer be paid twice

  • Refund tokens the mint reports already spent are dropped, so they no longer linger in storage and inflate the held balance

  • Mint recovery probes now hold their deadline even when fetch ignores the abort signal, so a down mint can no longer hold the recovery gate open

  • Model and upstream-path cooldowns are now exponential: a qualifying failure cools its own scope for 10s, doubling per consecutive failure up to 10 minutes. Streaks survive cooldown expiry and restarts, routstrd cooldowns --reset clears them, and the CLI reports expiries, path scope and failure streaks

  • Model-path resolution is spelling-aware: node alias_ids are retained and paths are merged across every spelling of a model, so a node advertising only a variant spelling is no longer dropped from routing. Added PPQ, Venice and Wafer DeepSeek routes

  • Auto-pin failover now degrades to ordinary provider routing instead of exhausting the request; per-model path lookups are fetched once per spelling and cached, and each node gets a 6s ranking budget so one slow node's oversized path payload can no longer make every request pay its timeout

  • Provider enable/disable reports the policy that actually took effect, backed by a two-tier Nostr review-label policy (default-enabled labels vs explicit opt-in)

  • Unreachable providers can no longer stall discovery: provider fetches get an independent deadline (~3s instead of up to ~136s), model-path lookups are bounded, the four relay filters refresh concurrently, and unchanged model shards are skipped during incremental discovery

  • Usage windows are now timezone-correct (IANA zone with DST-aware calendar arithmetic) and aggregate in SQL, cutting 7d over a 20k-row database from ~570ms to ~120ms

  • Added time-window filtering across the TUI, including a window-driven History tab with a dual-axis spend/token chart

  • The CLI prompts to update routstrd on start and restart, backed by daemon-side update state shared by the CLI and TUI, and added a rout alias for the binary

  • Startup no longer repeats discovery work, and onboarding waits for a non-empty model list and warns instead of aborting when an integration installer fails

  • Updated bundled routstrd to 0.4.17 and routstr-sdk to 0.4.12