omadia public preview — feedback welcome

[Weegy]

👋 Welcome — especially if you came here from the write-up on dev.to (or X / LinkedIn / Reddit / HN).

omadia is an open-source (MIT), self-hostable runtime for composing AI agents out of plugins — channels (Teams, Telegram, web chat), integrations (Microsoft 365, Odoo, Confluence, …), and capability providers (knowledge graph, embeddings, image generation, web search). It's single-tenant: run it on Docker Compose or a fly.io app, bring your own LLM key, and keep all the data on your side.

It's pre-1.0 / public preview — it works and we run it in production, but schemas can still break between minor versions. I'd rather get the hard design questions right with people who actually build agents than guess in private.

The two things I'd most like your take on

1. How would you scope memory between orchestrators?
   Today each agent/orchestrator gets a strict own memory namespace + knowledge-graph slice, so context can't bleed between them. It's clean, but it makes deliberate cross-agent sharing the awkward case. Where do you draw that line — per-agent, per-team, capability-scoped, something else?

2. Do you sanitise tool output before or after record expansion?
   We treat tool results as an untrusted boundary (a "summary + details" blob can carry PII into a context that shouldn't see it). Right now we expand the blob per record first, then gate per field. Do you sanitise before expansion, after, or at the data-source layer? What's held up for you at scale?

Other ways to help

• 🐛 Found a rough edge? Open an issue.
• 🧩 Which integration or channel would make omadia useful for you?
• ⭐ If you want to follow the public preview, star the repo — it genuinely helps us gauge interest.

Thanks for taking a look — I'll be around in this thread to answer anything.