A teammate that already knows the job.
Lifecycle-aware. Role-scoped. Voice + chat. Cites every source. APO lives inside the systems your business runs on — POS, ERP, HR — and handles the parts your operators don't want to do twice. Knows the new hire is new. Takes over when they need it. Learns who you are so it doesn't have to ask twice.
Two engines. One conversation.
APO is half deterministic flow controller, half generative LLM. The boring half is fast and free. The interesting half is grounded in your live operational data.
Deterministic where it should be
The ok / next / done handoffs run server-side as state machines. No LLM call, no token spend, no latency tax. The boring half stays boring.
Generative where it should be
Free-text questions hit the model, grounded in your live POS, ERP and HR. APO never makes up a number it can't trace back to a row. LLM provider is pluggable — default Gemini, swap NVIDIA / OpenAI-compatible runtimes, or bring your own.
Cited sources, every answer
Every claim links back to the entity it came from — order #, SKU, employee record, ticket. Operators get the receipts; auditors get the trail.
Voice that doesn't suck.
Streaming STT
Streaming transcription with on-device VAD. Knows when the operator paused for breath vs. finished a thought. Handles code-switching mid-sentence. Plug in any compatible STT engine — Whisper is just our default.
Realtime TTS
Small enough to run hot at the edge. Big enough to sound like a person, not a kiosk. Six voice families, all trained on operator-tone scripts. Swap to any TTS engine you prefer.
Server-side flow controller
The LLM never has to decide whether to ship a confirmation packet. State transitions are owned by Apogee. Voice just becomes another input modality.
Different role, different agent.
Put APO on one lifecycle for 30 days. Decide after.
Bring one workflow you'd rather not do anymore. We wire APO into your POS/ERP/HR, scope it to a single role, and let your operators run it for a month. If it doesn't earn its keep, we unplug it.