Inference infrastructure
Inference is electricity. We sell it like a utility.
Your agents run on your own APG appliance. The one thing they need that you may not want to host is the model itself. Solutions Factory runs that as a utility — metered, itemized, and dialable from fully managed to fully sovereign. Three tiers, one config line between them.
An electric utility buys power wholesale and bills you for what you draw, with the rate printed on the bill. We do the same with inference. You see the wholesale cost, you see our margin, and you see your total. If you would rather buy direct or generate your own, the meter comes out and you pay us nothing for inference. That is the whole model, and the rest of this page is just the three ways to wire it.
Tier 1 · SF default
Metered through Solutions Factory, itemized like a power bill.
Your APG appliance routes inference calls through the Solutions Factory proxy. We buy Anthropic capacity wholesale and bill you at a transparent retail rate. Prompt content transits under a zero-data-retention posture; your persistent data never passes through us. This is the default, and the path of least friction for a new customer.
Inference — monthly
IllustrativeThe margin is a line you can read, not a markup buried in a plan. You can watch your usage, forecast it, and compare it against bringing your own key at any time.
- Recurring and predictable. Pay for the tokens you draw, billed monthly.
- Zero-data-retention proxy. Prompt content is not retained; persistent data never transits SF.
- No commitment to stay. Move to BYOK or sovereign with one config edit.
Tier 2 · BYOK
Bring your own key. We charge for software only.
If you already hold an Anthropic contract or buy at enterprise volume, point APG at your own key. Anthropic bills you directly, Solutions Factory takes zero inference margin, and you pay us for the software license and support alone.
$ apg config set inference.tier byok $ apg config set inference.anthropic_key $ANTHROPIC_API_KEY inference: byok · billing direct to Anthropic · SF margin: none
Tier 3 · Sovereign
Self-host the model. Pay nobody for inference.
For the highest-privacy verticals — legal, healthcare, defense-adjacent — and operators with a GPU budget, run a local model on your own hardware. No inference cost, no SF margin, no prompt content leaving your network. A software-only sale.
Hardware, honestly. Run a 7B-or-larger model at Q4 quantization or better for reliable structured output. A single modern GPU with 12–16 GB of VRAM handles an 8B model at Q4 comfortably; larger models and higher quantization want more memory. Below that, structured-output reliability drops, and APG's validation will tell you so rather than letting it pass.
$ apg config set inference.tier sovereign $ apg config set inference.ollama_model llama3.1:8b-instruct-q4_K_M inference: sovereign · local ollama · SF margin: none
Which tier
Pick by profile, change your mind anytime.
| Your situation | Tier | Why |
|---|---|---|
| New customer, want the least friction | Tier 1 · SF default | Nothing to procure. Metered, itemized, predictable. |
| Existing Anthropic contract or enterprise volume | Tier 2 · BYOK | Use the rate you already negotiated; pay SF for software only. |
| Highest-privacy vertical or own GPU budget | Tier 3 · Sovereign | No prompt content leaves your network; no inference cost at all. |
See the numbers
Software and inference, priced separately and plainly.
The pricing page lays out the software-license tiers against the three inference tiers, with no per-seat line anywhere on it.