# Data Intel

Turn a complex research question into a cited report with a visible plan, traceable claims, and explicit checks for gaps and contradictions.

Status: active

## Data Intel turns research questions into cited reports

You approve the questions and source scope before collection begins. Data Intel then traces findings to source text, shows where evidence is missing or contradictory, and writes a cited report. You can inspect the evidence behind an answer before using it.

The local dashboard lets you create a run, review its plan and inputs, follow progress, and read the outputs. The command line runs each stage. The dashboard shows the command to use when a stage needs execution or a host model response.

## Six stages separate collection, analysis, and writing

Six stages separate source collection from interpretation and writing:

1. **Plan:** Turn the request into a decision question and a set of subquestions. Work starts after you approve the plan.
2. **Collect:** Gather sources within the approved scope from filings, web pages, local files, or prior research.
3. **Assess:** Score sources and extract claims with links to the exact text that supports them.
4. **Link:** Connect people, organizations, metrics, and claims in a research graph that can carry context across runs.
5. **Insight:** Identify unanswered questions, conflicting values, and trends before drafting conclusions.
6. **Write:** Produce a long report, a short version, highlights, and a presentation outline with citations.

## Data Intel checks evidence before writing

Data Intel attaches each extracted claim to its source and exact text span. The link stage connects claims and entities in a research graph. The insight stage finds unanswered questions, conflicting values, and trends before the model drafts a report. The writer uses citation markers assigned to those claims. A final check flags uncited numbers and broken markers; the coordinator leaves a run incomplete when those checks fail.

## The CLI executes stages while the dashboard handles review

The CLI executes each stage and owns its files while it runs. The dashboard lets you create a run, review inputs, approve a plan, follow progress, and inspect outputs. When a stage needs execution, the dashboard shows the command instead of running it. Data Intel keeps the plan, sources, stage outputs, and event history in local run files, so you can trace what happened.

## Local-only mode controls model access

Data Intel's model router rejects cloud calls by default. You can use a local endpoint or explicitly allow an approved provider. Data Intel binds the dashboard to loopback by default. Because the dashboard has no sign-in, it serves one user on the same machine.

## Links

- [HTML project page](https://rosslabs.ai/projects/data-intel/)