AI applications
The hard part is not the model. It's the approval.
Any provider will sell you inference. Getting an AI feature past a security review needs guardrails, spend limits, data residency, and an audit trail. Those are built in here.
The demo is easy. The rollout is not.
A prototype takes an afternoon. Then the questions start. Where does the data go? What stops it leaking personal information? What happens when someone loops it and burns the budget? Who used it, and can you prove it?
For an Australian government or enterprise client, those questions decide whether the feature ships. Answering them usually means building a governance layer nobody scoped — request logging, spend tracking, content filtering, an approvals story.
Here that layer already exists, and it applies whether you call the API directly or use the agents, skills, and workflows on top of it.
Built in
Answers, not action items
- Where does the data go?
28 of 30 models run with Australian-resident inference. On QuantGov, you see only those 28. The AI platform is in scope of our IRAP assessment.
- What stops a leak?
Guardrail presets detect personal information and filter harmful content. Custom filter policies catch whatever else matters to you.
- What caps the cost?
Spend limits checked before the request runs — per organisation, per user, per API token, per interface.
- Who used it?
An audit record for every chat, completion, embedding, and AI Search request under Object Lock, with the user, the token, the model, and hashes of the prompt and response.
Getting here
Change the base URL
There is an OpenAI-compatible endpoint for chat completions, embeddings, and model listing. If your application already speaks that API, pointing it here is a configuration change rather than a rewrite.
When you want more than inference, the agents, skills, tools, and workflows are waiting on the same platform under the same controls.
What teams build here
The governance is the same whichever of these you start with.
Answers from your documents
Vector collections with automatic chunking and semantic search, so responses cite your material rather than the model guessing.
- Automatic chunking
- Source attribution
- Metadata filtering
Search that understands intent
A managed search product with crawling, indexing, semantic retrieval, and grounded chat you can embed in a page.
- Semantic and keyword
- Automatic crawling
- Embeddable widget
Assistants in your product
Multi-turn chat with session memory and streaming, over REST, server-sent events, or WebSocket.
- Session memory
- Streaming
- Context compaction
Automation behind the scenes
A webhook starts a workflow, agents do the work, and the result lands where you need it. Pause for approval when it matters.
- Webhook triggered
- Human approval steps
- Versioned
Consistent behaviour across features
Put your standards in a skill and assign it to your agents, instead of copying the same instructions into each prompt.
- Import or author
- Pin GitHub skills
- Shared across agents
A feature that passes review
Mandatory guardrails, model allowlists, spend ceilings, and an audit log you can hand to a client.
- Mandatory presets
- Pre-request caps
- Immutable audit
Build the feature, not the governance layer
Start with a free trial and talk to our team about your security requirements.