ZeroTrace OSINT
AI Assistant
What the built-in assistant can do, how it runs OSINT's own tools with full visibility, the local and hosted model providers it supports, and the guardrails that keep it honest.
AI Assistant
ZeroTrace OSINT includes a built-in AI assistant that works alongside you inside the app. You can ask it to look something up, and it reaches for the same tools you would, runs them, and brings back the results, then helps you understand what they mean.
What it does for you
- Runs tools on your behalf — describe what you want to know and the assistant picks and runs the right tools from the library, so you don't have to chain them by hand.
- Shows its work — every tool call is visible in the conversation: what it called, the arguments it passed, the result that came back, and how long it took. Nothing happens off-screen, so you can check its reasoning and trust the result.
- Helps you read results — it can explain what a lookup turned up, connect findings across tools, and turn a wall of output into something you can act on.
- Files findings for you — when it's confident a result belongs in your case, it can save it as a finding directly, the same way filing one by hand does.
It works through the same tool library you have, with the same network/live/local labeling and the same authorized-use expectations. It can only do what the workbench can already do; it just does it faster and explains as it goes.
Model providers
| Provider | Note |
|---|---|
| Ollama | Runs on this machine. Nothing leaves it. |
| LM Studio | Talks to a local server on port 1234. |
| llama.cpp | Talks to a local llama-server endpoint. |
Local providers are the default for a reason: an investigation tool's own conversation shouldn't have to leave the machine it's running on. Point the assistant at any of the three and it stays entirely on-device.
Switch providers and models from the assistant's picker at any time; your key is stored the same way as any other tool credential — see Privacy and Settings.
Guardrails
Every claim the assistant makes gets checked against the tool output it actually produced. A finding it writes into a case carries a grounding verdict — grounded, partial, unsupported, or not-checkable — recording how much of it was backed by a real tool call versus inferred. That verdict travels with the finding into the dossier, so a report never presents the assistant's inference with the same weight as a lookup it actually ran.
For hosted providers, which bill per token, usage is tracked per provider and model so you can see what a session is costing before it surprises you.
The assistant helps you investigate; the responsibility for what you look into stays yours. The same authorized-use rules that apply to running a tool by hand apply to anything you do with the assistant's help.