Run locally.
Connect a local Ollama installation or another supported local endpoint. Keep inference on infrastructure you control.
Ghost Observatory / AI integration
Your analysis does not need an AI landlord.
Coming soon
Choose a local model, an existing provider or your own inference service.
Configure the endpoint and credentials, and decide which published records it may read. Ghost Observatory supports OpenAI-compatible endpoints, local Ollama and custom connections through YAML configuration or an adapter.
No Ghost Citadel account. No mandatory subscription. No model hosting on our side.
No data collection by Ghost Citadel.
Your keys. Your model. Your data.System walkthrough / illustrative flow
01 / Source recordsCustomer-controlled inputs
02 / Declared rulesYAML · authority · admission
03 / Deterministic engineIdentity · sequence · intervals
04 / Result fileTraceable SQLite output
Optional AI advisorExplanation · mapping proposals · questions
The engine processes declared source records and produces an authoritative result. No language model is required or called by the calculation core.
Diagram labels remain available without graphics acceleration.
Deterministic at the core
The calculations don’t answer to a language model.
Identity resolution, sequence admission, interval mathematics and configuration enforcement remain in the deterministic engine. These calculations never call a language model. AI sits outside that boundary as an optional advisor; it cannot replace the engine's declared semantics or authoritative results.
Model connections
Choose how the model connects.
Bring an API.
Use an OpenAI-compatible endpoint with your own credentials and provider arrangement.
Host it yourself.
Connect an internal inference service through a thin adapter. Your deployment, your rules.
Optional advice
Three uses for an advisor.
Why was this trace withheld?
Ask the advisor to explain an admission decision already reported by the engine.
Suggest a field mapping.
Review a candidate YAML mapping. The engine still validates every declared requirement.
Show repeated handovers.
Translate a natural-language request into an operation the existing software understands.
Your data stays yours
You control the connection and the data.
Ghost Observatory does not send your operational data, analyses or model conversations to Ghost Citadel. There is no Ghost Citadel telemetry in the engine.
If you choose a hosted AI service, the information you send to that service is governed by your chosen provider and your configuration. Use a local model when you want inference to remain on your own infrastructure.
There is no Ghost Citadel model relay.
The core runs on its own.
Leave the adapter unconfigured and the engine operates as before.
Explore Ghost ObservatoryInspect the data boundary before connecting a model.
The result file and Python API have different access boundaries. Choose the tables and records an advisor may read, then review the provider and deployment configuration.