Search · For: agent

Complex questions need more than one symbol search, but free-form search loops waste tokens.

Deep Search explores code iteratively, records evidence and stops when confidence or budget reaches the threshold.

$ aindex deep-search "where is session state invalidated?" --budget 12000 --json

steps
evidence
citations
stop_condition
contract_verified

Command

Deep Search

Iterative local research with evidence, citations and a stop condition.

CLI

aindex deep-search "where is session state invalidated?" --budget 12000 --json

MCP

context action=deep-search

Example output

What the Brain does automatically

Complex code questions become budgeted research with citations instead of another open-ended search loop.

intent

When you are still mapping the codebase.

intent

Fires on intent (research).

−69.1%Proof

MeasuredMedian token reduction across 7 real codebases (1,157 files).

Sourcepublic/assets/benchmarks.json

AvailableRuntime feature status · 0 evidence item(s). Metadata-only; no raw source included.

Required tier

Answers a cross-repo architecture question with cited files and a budget guard instead of an unbounded search loop.

Advanced · €9.99/mo

Visible as a product workflow and pilot proof target; do not treat as fully shipped enterprise automation yet.

Buyer grade

Names the parity workflow and proof path without claiming full Sourcegraph-class depth.

Visible in

This workflow stays discoverable across setup, CLI, MCP status, docs, dashboard and pricing — never a hidden command.

Feature visibility checklist

This keeps grouped product names discoverable without claiming a separate maturity level from the underlying workflow.

Deep SearchFoundation

Budgeted multi-step research returns citations and a stop condition.

Budgeted research