cloud.abridge.services

Getting started

Updated 2026-08-29

There's nothing to install to use APEX — it lives in Slack. Cloning the repo is only needed if you want to run or extend the agent locally.

Ask APEX in Slack

  1. In any channel, DM, group DM, or thread, mention @APEX with your question.
  2. APEX reads the thread context, adds an hourglass reaction while it works, and replies in-thread.
  3. When it's done, the hourglass becomes a checkmark. If confidence is below the triage threshold, the answer is flagged for a human.

Tips for good answers:

  • Ask about current state ("is encounter-processor erroring in prod right now?") — that's where live-source-first shines.
  • Point at a specific service, workflow, dashboard, or PR when you can.
  • APEX creates PRs, tickets, or docs only when you explicitly ask.

APEX only responds to a real @APEX mention — quoting the word "APEX" won't trigger it, and external users are blocked.

Run or extend APEX locally

Clone abridgeai/apex and open it in an agentic IDE — the repo rules already point agents to agents/DIRECTIVE.md.

Python dependencies route through the python-virtual Artifact Registry proxy, so uv needs a short-lived gcloud token. The Makefile targets fetch it for you (you need membership in one of abridge-eng-gcp-access, appteam-all, or ml-team):

make uv-sync
pnpm install
scripts/uv_with_auth.sh run pre-commit install

The fr developer CLI

APEX ships an internal fr CLI (also the local entrypoint for Frankie builds). Run it through the authenticated uv wrapper from the repo root:

# discover commands
scripts/uv_with_auth.sh run fr --help

# check local Apex Console prerequisites, then start the stack
scripts/uv_with_auth.sh run fr dashboard doctor
scripts/uv_with_auth.sh run fr dashboard start

# check Frankie build prerequisites without mutating GitHub or the DB
scripts/uv_with_auth.sh run fr build doctor

Configure PHI-safe behavior

If you extend APEX, keep it inside the HIPAA boundary: aggregate BigQuery output only, never post raw clinical text or identifiers, and read agents/DIRECTIVE.md for the complete rules on evidence tiers, environment awareness, Slack attribution, response length, and safety.