Sanitisation
Hosomaki's sanitisation layer is the first and mandatory step of every pipeline invocation.
What gets scrubbed
| Category | Examples |
|---|---|
| IPv4 addresses | 192.168.1.1, 10.0.0.254 |
| IPv6 addresses | ::1, fe80::1 |
| Hostnames and FQDNs | myserver.internal, db.prod.example.com |
| Filesystem paths | /home/alice/, /etc/secrets/ |
| UUIDs | 550e8400-e29b-41d4-a716-446655440000 |
| Usernames | As appearing in log lines |
| Credentials and tokens | API keys, passwords in log output |
Architectural position
Sanitisation happens before the prompt package is called:
go
san := sanitiser.Default()
sanitisedLogs := san.Sanitise(rawLogs)
generationPrompt := prompt.Explain(prompt.ExplainInput{
Logs: sanitisedLogs,
// ...
})The prompt package always receives pre-sanitised data and has no knowledge of what was stripped.
Per-line sanitisation
For streaming use cases like watch, sanitiser.DefaultPerLine() applies the same scrubbing rules to each journal line individually as it arrives from the tail, before it enters the batch buffer.
What sanitisation does not do
- It does not anonymise log structure or timing information
- It does not prevent the model from inferring general system characteristics
- It is not a substitute for ensuring your Ollama instance is secured at the network level
Privacy guarantee
Because Ollama runs entirely on your machine, and the sanitisation layer removes identifiable data before prompting, Hosomaki provides a strong practical privacy guarantee: no sensitive system data leaves your machine.
See Data Privacy for the full data handling policy.