Compute Units
Compute Units (CUs) are Alphonce's single meter for AI work — research, agent turns, market pulls, and execution signals. One currency across the product, so you are not juggling separate token bills per feature.
What is a Compute Unit?
A CU is a published cost weight for a unit of work. Light lookups cost a little; a full research cycle costs more. Your plan includes a monthly pool of CUs. When a feature runs, Alphonce deducts the published weight (or a reconciled actual after the call).
Non-refundable once used
Published weights
These are the user-facing conversion rates used across the product:
| Task | Cost | Typical use |
|---|---|---|
| 1 market-data pull | 1 CU | Quotes, history bars, cached lookups |
| 1 trade execution signal | 10 CU | Order routing / signal work |
| 1 agentic thinking step | 50 CU | A chat turn or agent reasoning step |
| 1 full research cycle | 500 CU | Deep research / multi-stage synthesis |
Examples: an agent chat turn is typically one thinking step (50 CUs). A deep research pass is one full cycle (500 CUs). MCP market reads are priced like a market-data pull.
Plan quotas
Each plan sets a weekly limit — the allowance worth planning against — plus a smaller session window that refills every two hours so one busy afternoon cannot spend the whole week at once. Hitting the session window is a short wait; hitting the weekly limit means waiting for the reset or moving up a plan. When either is empty, metered calls stop, unless you have turned overage on in settings.
Agents run at 40% of the interactive rate. An unattended decision is one model call over a fixed market snapshot — no tool loop, no retrieval, no growing conversation — so it costs less to serve and is priced that way: 20 CU per tick against 50 for a chat turn. A bot deciding once each weekday spends roughly 100 CU a week. Talking to a bot is charged at the normal rate, because a human in the loop means the expensive path.
| Plan | CUs / week | Session (2h) | Requests / min | Overage rate |
|---|---|---|---|---|
| Free | 3,000 | 500 | 20 | Not available |
| Plus | 20,000 | 2,500 | 60 | Optional · 0.50¢ / CU if enabled |
| Premium (Chui) | 50,000 | 6,000 | 150 | Optional · 0.45¢ / CU if enabled |
| Pro | 130,000 | 15,000 | 300 | Optional · 0.40¢ / CU if enabled |
| Team | 450,000 | 50,000 | 1,000 | Optional · 0.30¢ / CU if enabled |
| Enterprise | 1,200,000 | 125,000 | 1,000 | Optional · 0.30¢ / CU if enabled |
Overage is off by default
How spend works
- 1Pre-flight. Before a metered call runs, Alphonce checks your tier, remaining CUs, and rate limits.
- 2Execute. The feature runs (research, agent turn, MCP tool, and so on).
- 3Deduct. The call cost is charged against your monthly balance. If the pool is empty and overage is not enabled in settings, the call is rejected.
You get spend alerts as you approach the top of the pool (50%, 80%, 95%, depleted).
MCP and the API
Connecting your own client over MCP is included from Premium (Chui) upwards. Tools that pull data or run research still spend CUs from the same pool as the app. You can inspect usage with the GET_COMPUTE_USAGE MCP tool when your token allows it.
The API reference covers tokens and transport only — metering stays in this CU model.
Bring your own model
On eligible plans, the BYO LLM add-on lets you point Alphonce at your own provider key. You pay that provider for tokens; Alphonce still meters platform work (data, orchestration, gated tools) in CUs. Inference on your key is not double-billed as Alphonce CU burn.
See it in the product