Routstr is a decentralized AI inference router built on Nostr and Cashu. Providers announce available models, capabilities, and prices on the Nostr network using standard relays; clients discover providers through those relays and pay per request using Cashu ecash, a Chaumian e-cash protocol denominated in Bitcoin sats. Settlement is instant and the payment tokens are unlinkable from the inference request, so the marketplace has no central operator, no account creation, and no card-on-file requirement. Providers can also accept BOLT12 Lightning payments alongside Cashu notes.
The architecture separates three concerns: Nostr handles censorship-resistant discovery and metadata (provider listings, model menus, pricing), Cashu handles the payment, and the inference itself is provided by whoever the client selects. The open-source otrta-client is the reference payment gateway (github.com/Routstr/otrta-client) and exposes an OpenAI-compatible API, so existing client code that targets the OpenAI SDK can route requests through Routstr with a base-URL change. Documentation lives at docs.routstr.com.
The project was built during Sovereign Engineering cohort SEC-05 in Madeira and is maintained by the Sovereign Engineering team and Routstr contributors. The design intent is censorship resistance: when a national government or a corporate provider blocks an activist's access to a particular model, Routstr allows continued inference by routing through other participants who still have access to it. The marketplace structure prices that residual capacity instead of relying on volunteer donation of accounts.
HRF awarded Routstr in its November 24, 2025 AI for Individual Rights round. The Cashu-plus-Nostr combination is the same primitive stack used by other freedom-tech projects in the Bitcoin space; Routstr is the most developed application of it to the AI-inference layer. The competing alternatives are direct API access (KYC and geographic block exposure) or VPN plus single-account hopping (slow and account-suspension prone). The structural argument is that pseudonymous market access to inference is the only design that survives both provider-level deplatforming and state-level network filtering, because no single party holds the routing table.
Recipient
Sovereign Engineering / Routstr team
Funder
Human Rights Foundation · foundation · Global
Funds open-source AI tools that put inference, agentic capability, and private compute into the hands of dissidents and civil society under authoritarian regimes.
Primary source
https://hrf.org/latest/hrf-announces-support-for-projects-advancing-ai-for-individual-rights/
Additional sources
More from Human Rights Foundation
- Stringer Safety 2026-01-18 · Hackathon prize
Safety app for journalists in high-risk environments. Reporters maintain a trusted contact network with shared location and status; a one-click SOS triggers response recommendations using the encrypted Maple AI assistant.
- Pathos 2026-01-18 · Hackathon prize
Application running on the decentralized Nostr protocol so it cannot be shut down by governments. Built at HRF's first AI Hack for Freedom in Austin, January 2026.
- Maple AI 2025-11-24 · Undisclosed
Open-source, end-to-end encrypted AI assistant built by OpenSecret using secure enclaves and confidential computing. Activists in authoritarian environments can use frontier LLMs without their queries being scanned, stored, or handed to governments. HRF support funds capability improvements and new features for activist research workflows.
- OpenCode 2025-11-24 · Undisclosed
Fully open-source agentic coding platform that runs entirely locally. Users can inspect every line of code, avoid surveillance, and build software without routing sensitive work through corporate infrastructure, making it usable for civil-society teams under authoritarian rule.
- PlebDevs AI Development Course 2025-11-24 · Undisclosed
Developer education platform focused on Bitcoin and open-source technology launching an AI development course for beginners building open-source AI tools in repressive environments.