Why your team’s AI should always show its sources

There’s a version of “AI for your team” that’s genuinely dangerous: a confident chatbot that summarizes your company’s knowledge and gets it subtly wrong. No sources. No way to check. Just vibes, delivered fluently.

We didn’t want to build that.

Two rules

When we designed Halo, the AI inside Ollasync, we started from two constraints:

1. Every answer cites its sources. If Halo tells you a decision was made, you can click straight to the message where it was made.
2. Halo only sees what you see. It answers strictly from the channels you’re a member of. Private channels stay private — the AI never becomes a backdoor around your permissions.

Why grounding beats fluency

A grounded answer does three things a generic chatbot can’t:

• It’s verifiable. You don’t have to trust Halo — you can check it in one click.
• It’s current. It reflects what your team actually said, not a stale training snapshot.
• It’s safe. Because it’s built from your messages and respects access, it can’t invent a “fact” your team never established.

Knowledge that compounds

The nice side effect: as your team talks, the graph gets richer. Meeting transcripts flow in. Decisions get captured the moment they’re made. Six months later, the answer to “why did we choose Postgres?” is still one question away — with the receipts.

That’s the bar we think team AI should clear. Anything less is just a confident guess.

See how Halo works →