The concern is understandable. General AI chatbots charge by the token, and a compliance engagement that runs hundreds of prompts — reading policies, drafting answers, mapping controls — could produce a notable bill. The question is whether that model applies to purpose-built compliance tools. For Compass, it does not.
The Per-Token Fear
The cost concern comes from a specific pattern: a general AI platform where every prompt costs something, complexity increases token usage, and the only way to limit spend is to limit use.
In that model, compliance work is expensive because compliance work is verbose. Reading a fifty-page policy, analysing its control coverage, and drafting a gap memo generates thousands of tokens of input and output. A team doing this for every audit cycle, every questionnaire, every framework crosswalk would face unpredictable and growing costs.
That model describes general-purpose chatbots used for compliance. It does not describe purpose-built compliance tools.
What BYOK Changes
Bring-your-own-key (BYOK) changes the cost equation at a structural level. Instead of the tool vendor charging a markup on every model call, the team pays the model provider directly for what they use — and the tool itself charges a flat seat fee.
| Cost model | What you pay | Who controls spend |
|---|---|---|
| Vendor-managed AI | Per-token or seat includes model margin | Vendor sets the rate |
| BYOK | Seat fee to tool + model key to chosen provider | Team chooses the model and rate |
| Local | Seat fee only | Team (no model cost) |
BYOK means the tool does not add a margin layer on top of the model cost. If a team already has an OpenAI or Anthropic account, the same key works. If the team negotiates a better rate through a cloud provider, that rate applies. The tool is not in the middle of the pricing.
Running Local: Zero Egress, Near-Zero Model Cost
Local models eliminate API costs entirely. Compass supports Ollama, LM Studio, and MLX — all run on the team's own hardware, with no per-token charge and no network egress.
The trade-off is performance. Local models are smaller and may produce lower-quality output on complex compliance tasks (cross-framework mapping, nuanced risk analysis) compared to frontier models like GPT-4 or Claude. For routine tasks — summarising a policy, drafting a memo from structured data, filling a standard questionnaire field — local models are often sufficient.
The practical approach is to use both: local models for everyday drafting and review, frontier models through BYOK for complex analysis, and switch between them per task without changing tools.
Cost Control Through Architecture Choice
Beyond the key and the model, the tool's architecture determines how many tokens a compliance task consumes.
A general chatbot treats every conversation as fresh context — the model re-reads the same documents each time you ask a related question. A purpose-built compliance tool maintains workspace context across sessions, so the model reads your evidence base once per agent run rather than once per prompt.
The result is that a compliance tool with workspace awareness uses fewer total tokens than a general chatbot doing the same task, even when using the same underlying model. The saving comes from architecture, not from limiting use.
FAQ
Do I need an expensive API key to use Compass?
No. Compass supports local models that run without any API key. For teams that want frontier-model quality, a standard OpenAI or Anthropic API key works — the same key you might already use for other tools. The cost is whatever the provider charges for that key, with no markup from Compass.
What if I use Compass all day — will costs add up?
With a flat seat fee and your own key, heavy usage does not increase the tool cost. The model cost depends on your provider's pricing, which is the same whether you use Compass or any other application with that key.
Can I switch between models day to day?
Yes. Compass is model-agnostic. You can switch providers, use different models for different tasks, or toggle between a frontier API and a local model without changing tools or workspaces.
Is there a hidden markup on AI usage?
No. BYOK means you pay the model provider directly. Compass charges a flat seat fee. There is no per-prompt, per-token, or per-agent-run margin added to your model costs.
The cost concern around AI compliance is real — but it applies to general chatbots, not purpose-built tools with BYOK and local-model support. Compass by Truvara uses a flat seat model with no vendor markup on AI usage: bring your own key for frontier models, or run local for zero inference cost. You control the spend. Join the waitlist.