AI apps love to hold your files
Upload a logo, a video, a PDF, and it lives in their database or their bucket. You end up renting access to your own bytes — and paying for the privilege when the file is a 200 MB render.
eidan v0.20.0 takes a different line: your files go to object storage you control.
Your bucket, your keys
Point eidan at an S3 bucket — or any S3-compatible store (Cloudflare R2, MinIO, Backblaze) — with your own credentials, and it offloads large files out of Postgres and into your bucket. Reads stream back through the engine, which holds the keys, so the web tier never touches your storage credentials. Nothing about your media leaves infrastructure you own.
Direct-to-bucket, past the 4.5 MB wall
Serverless platforms cap request bodies — about 4.5 MB on Vercel — which quietly kills video. eidan issues a presigned URL and the browser PUTs the file straight to your bucket. The bytes never pass through the app server, so a phone-shot clip or a screen recording uploads fine. This is the groundwork for content repurposing: get the raw media in cheaply, then let agents slice and publish it.
You decide where things land
A single Storage setting picks the policy — keep everything in the database, offload large files automatically, or offload everything — plus a toggle for direct browser uploads. No env-file surgery; the engine honours the setting on every upload.
Why it's built this way
eidan is a self-hosted, plugin-extensible agent OS. "Own your stack" has to mean your data too — not just the model weights or the prompt. Storage is infrastructure that makes self-hosting worth it, so it's AGPL core, not a paid add-on. The convenience layer — managed buckets, managed posting — is where a hosted tier could sit later; the capability ships open.
It's all in v0.20.0. Self-host it, and point it at your own bucket.