INTERNAL / EXTERNAL
The safe assumption is outside.
Self-hosted providers (vLLM, Ollama and kin) are INTERNAL. The named cloud labs are EXTERNAL. Anything we do not recognise is EXTERNAL. That default is in the data model, not a slide.
SW·004 - Compliance in the substrate
Automate the work.
Keep the receipts.
Your AI infrastructure with built-in compliance, not compliance bolted onto AI infrastructure. We lead with privacy because we believe in it, not because it scares people into buying. GDPR is a standard we meet. The EU AI Act is a set of obligations we walk through with you, against the work you actually run.
What we actually record
Two trails.
One picture.
Every model call writes an inference log: which model, which provider, how many tokens, latency, who triggered it, and, when the call came from a workflow, which execution and which node. When an asset is included in a prompt, a second record notes the asset, the provider, and whether that provider is internal or external to your infrastructure. Those two trails are how we can tell you what went where.
- 01 Model, provider, adapter, tokens in and out, cost, timestamps
- 02 User, chat, workflow execution and node when the call ran inside a process
- 03 Asset name, type, and whether the provider is INTERNAL or EXTERNAL
- 04 Unknown providers classified EXTERNAL by default, because that is the safe assumption
- 05 Human gates persist the person who decided, and when
The record, as we present it
One operation.
Readable.
The block on the right is the shape we show a visitor. The live admin view is the inference log, with the asset trail beside it. We are joining those into a single operation record. Until that ships, we will not pretend the columns are already one table.
AI_OPERATION_RECORD
#66a1d837
asset
quote_7731.json
workspace
—
workflow
—
model
claude-sonnet-4
provider
anthropic
classification
EXTERNAL
jurisdiction_exit
true
tokens_in / out
1,842 / 611
authorised_by
—
risk_class
unregistered - EU AI Act Annex III: n/a
retention
unretained
recorded_at
2026-09-22T12:40:05Z
exportable · queryable · yours
Exhibit pending
fig-01-inference-log.png
EU AI ACT
Walk through the obligations you have.
Impact-assessment templates sit in the platform so you can document the work against Annex III, where it applies, and say so plainly where it does not. We do not sell fear. We present the situation clearly.
NO LOCK-IN
We do not want your data.
Income comes from work you pay for because it improves the business. We do not mine what you store, sell it, or make leaving difficult. Export is standard. Retention is a policy, not a hostage.
Next step
Bring the process
everybody dreads.
We map it, run it on SynthWise, and hand you the receipts. If it works, the next process is mostly configuration.