fed_AI / fed_AI is on Nostr

fed_AI is a federated, pay-as-you-go AI inference network that routes requests across independent nodes. It focuses on low latency, competitive pricing, pluggable models, and privacy-preserving operation without centralised control.

nostr://p/npub1khf5amw8mrupe649a5mhkk3d2d8wjj06urz6kwawpftk4f68tjlstdmd3q

Last Notes

2026-01-22 22:41:25 UTC

Fed_AI progress report: cleaning up the core and polishing the UX. 🤖💨
Under the Hood:
- Refactored the massive 3k-line Router HTTP handler into specialized modules for health, ranking, and payments.
- Implemented persistent config volumes so your node identity survives restarts.
- Fixed 502/401 edge cases in Docker and added service auto-restarts.
Admin Dashboard:
- No more guessing! Added text search for GGUF models with suggestions and real-time download progress bars.
- Flexible Auth: use your favorite extension, a remote signer, or just paste an NSEC to claim your node.
- Model Management: You can now download and "Set Active" different models directly from the UI.
The bridge between #Nostr and AI is getting stronger.
#fed_ai #open_source #p2p #devlog #ai #nostrdevs #decentralized

2026-01-19 02:50:08 UTC

fed_AI update: streaming inference is live end‑to‑end (protocol/router/node/SDK), added worker threads for sig/validation offload, and refreshed the simple‑chat demo to use the SDK + new Groq model.
Lint/tests green. Try it out and poke holes #nostr #nostrdevs #fed_AI #AI #vibecoding #decentralisedAI #decentralised @nprofile…p7sp

2026-01-18 04:23:50 UTC

- reply

I found switching to a different coding AI can it work. Seems they can have different perspectives....

2026-01-16 22:51:38 UTC

Most AI today lives in one place. One company, one cloud, one bill.
fed_AI is an attempt to flip that around.
Instead of sending your request to a single massive server, fed_AI breaks it up and sends it across a network of smaller machines. Home PCs, small servers, even low-power hardware that would normally sit idle.
From a user’s point of view, it’s simple. You open a client, ask the AI to do something useful, and get an answer back. Writing, analysis, media generation, small tools. You pay a few sats over Lightning and move on.
What’s different is what you don’t see. There’s no account to sign up for, no single provider holding all your data, and no assumption that everything has to run in a huge data centre. The network figures out where the work should go based on what’s available and what makes sense at the time.
It’s early, and it’s experimental. But the idea is straightforward.
AI shouldn’t be a platform you rent. It should be a network you can participate in.
#fed_AI #nostr #decentralisedAI #lightning #p2p

2026-01-16 03:00:11 UTC

**fed_AI is live.**
fed_AI is an experiment in decentralised AI infrastructure.
The goal is simple: break large AI systems into composable, auditable services that can be run by independent operators, routed dynamically, and paid for on a pay-as-you-go basis.
Instead of one opaque model doing everything, fed_AI treats AI as a network:
* Routers decide where work goes
* Nodes specialise in narrow tasks
* Models are pluggable, replaceable, and swappable
* Capacity comes from many machines, not one provider
This is not about chasing scale for its own sake. It is about resilience, cost control, locality, and choice.
Key principles:
* Open source by default
* No data scraping or silent retention
* Deterministic routing and traceable execution
* Friendly to small operators and hobbyist hardware
* Works alongside existing proprietary and open models
Early focus is on proof-of-concept code, clear documentation, and a threat model that assumes untrusted participants.
This is experimental. Expect rough edges. Expect iteration.
If you are interested in decentralised systems, AI infra, or running useful services on ordinary machines, this project is for you.
More soon.
#fedAI #Nostr #DecentralisedAI #OpenSource #DistributedSystems