halley

Flagship · In development

An operating system for AI at work.

One place where a team's agents talk, listen and act. It fits whatever a team needs, not one industry.

Ask for early access

Four parts

01

Chat and voice, one memory

The same agent answers on the web, on the phone and by voice, and remembers the person across all of them. A conversation started in chat can finish on a call.

Related open sourceTring
02

Workflows you can read

Agents are designed as versioned workflows. Every run can be replayed, every change tested against real failures before it ships, and any version rolled back.

Related open sourceAgentReplaywitness
03

Small models for routine steps

Task-specific small models handle routing, extraction and checks, quickly and cheaply, so the large models are kept for the hard reasoning.

Related open sourceguardian
04

Integrations that act

Agents work through the tools a team already uses, with identity and permissions they can prove, so they can take actions and not only talk about them.

Related open sourcePact

Where it stands.

It is in development. We are building it in the open where we can, shipping the pieces as open source first, and opening early access to a small number of teams who want to shape it.