Zum Hauptinhalt springen

Discovery & Analysis

Discovery & Analysis is the default project mode. nori works in two phases: it first builds a local source index of your codebase (called the Wiki in versions before 1.26.0), then evaluates your compliance policy against that curated context. For how the two-phase pipeline works under the hood, see Architecture.

Index first

In this mode the source index must be built before controls can be evaluated — the analysis draws on the indexed knowledge base as its evidence. Build (or refresh) the source index first, then run the analysis.

1. Build the source index​

This step is required before any controls can run. Click Build the source index at the top of the project page to build the source index topics for your codebase; these runs are tracked under Source index runs. Once the source index is in place, its context is injected into the control evaluations in the next step.

Optimization: Concise source index

You can enable the Concise source index setting. This instructs the indexer to generate shorter, fact-dense source index pages, capping verbosity. Because source index content is injected into downstream control evaluations, a concise source index drastically reduces overall analysis token costs.

Local Wiki Builder

2. (Optional) Tune the analysis​

Before you run, you can adjust behaviour under Project Settings → AI & Analysis Config ("Adjust inference models and scanning logic"):

Analysis Configuration

  • Analysis Depth — Light (Surface-level scan), Medium (Balanced details), or Extreme (Deep semantic traces).
  • Generate Executive Summaries — produce a high-level summary alongside the findings.
  • Performance & Parallelism — Analysis Parallelism (max concurrent LLM calls for control evaluation), Source index parallelism, Max Objectives per Agent, and Agent Timeout (seconds).
  • Budget Limits (Claude CLI) — Analysis Max Budget (USD) and Source index max budget (USD) set per-agent spending caps; leave empty to disable the cap.
  • Resumability & Optimization — Skip Existing (Analysis) ("Resume runs by skipping already-evaluated controls"), Skip Existing (Source index), and Compress source index context ("Strips filler prose from the source index before it is sent to the model. How much it saves depends on how much prose your source index holds — tables, paths and code barely compress."). No fixed saving is promised, because there isn't one: instead, each run measures what compression actually removed and nori shows that figure under the toggle — "Saved x.x% of source index characters on the last run", or "No measurement yet" whenever the most recent run measured nothing. The run log prints the same figure for that run.

3. Run the analysis​

With the source index built, open the project and, from the project header, click Run analysis. In the configuration modal that appears, you can customize the run options or select a different AI model just for this run (overriding the default project models).

Once started, nori dispatches a bounded sub-agent per control and evaluates your source code against the selected policy, using the source index as context. Progress streams live on the Controls tab as each control executes.

Controls executing during a run

The project dashboard organizes everything into tabs: General, Controls, Source index, Document index, Evidence, and Configuration. Past runs are listed under Analysis runs.

Add external evidence​

Some controls need proof that isn't in your source code. You can attach documents and index them for use during analysis — see External Evidence.

When your run finishes, continue to Review Results.