Deep Search is the orchestrated set of AI agents Relm runs against a property to gather and structure the information you'd otherwise spend a week assembling by hand. It's the engine behind every section card on the property workspace.
(A few in-app buttons and the codebase still say "Deep Research" or "DeepSearch" — same thing.)
What it does
When you start Deep Search, more than a dozen specialized agents launch in parallel:
- One agent goes after tax history from public records and an institutional property graph.
- One assembles comps from listing platforms and ranks them by similarity.
- One gathers points of interest from a maps and places API.
- One compiles demographics and area stats from federal demographic surveys.
- One builds the unit-by-unit table using an agentic crawler with independent page-fetch verification.
- One pulls submarket trends and supply pipeline for Market Analysis.
- One extracts and indexes any uploaded documents.
- An AI Summary agent runs at the end, synthesizing everything into the executive narrative.
- A Financial agent builds the 10-year pro-forma. On a full Deep Search, financial analysis auto-runs after the research sections finish (you can still regenerate later if you change assumptions).
Each agent writes its results into the property state with citations attached. As each finishes, the corresponding section card flips from Pending to Done.
How long it takes
A typical Deep Search run is 5–10 minutes for a multifamily property in a major metro. Tertiary markets, asset classes with thinner data coverage (industrial, certain retail), or properties with very large rent rolls (1000+ units) can take longer. When financials auto-run (or you regenerate), pro-forma generation adds another 7–20 minutes.
What you can do during a run
- Leave the page — the run continues server-side. You can come back to a fully-populated workspace.
- Read other sections — already-completed sections are fully interactive while later sections are still running.
- Use the chat panel — chat falls back to read-only on sections that aren't done yet.
- Cancel a section — click cancel on any in-flight section card.
- Cancel the whole run — from the property header.
You can't edit a section while its agent is actively writing to it (the lock prevents conflicts). Once the section flips to Done, editing is unblocked.
Re-running
You can re-run any individual section or all sections at any time. Re-running a section overwrites its data — your manual edits to that section will be discarded unless you've saved them as overrides.
The first full Deep Search auto-starts financial analysis after research completes. Later pro-forma regenerations are an explicit action (each generation uses credits on Self-Serve). Re-running research sections alone does not auto-regenerate the pro-forma — regenerate from Financial when your assumptions changed.
When to run it
Recommended patterns:
- Instant Preview, then add. Search an address on
/dash, review the free Instant Preview (Street View, facts, nearby), then Add to portfolio when you want a full workspace. - Address only. Add a property by address, click Start Deep Search, watch it populate — financials follow automatically. Fastest for top-of-funnel screening.
- Documents first, then research. Add the property, upload your rent roll/OM/P&L, then start Deep Search. Documents are picked up as a primary source — the rent roll wins on units, the P&L feeds expense baselines, the OM contributes context to the AI Summary.
On Deep Search you can also pick an AI provider when your org has multiple options enabled — use the provider control on the run button.