Leni

Where Leni Fits

Most AI tools aimed at real estate teams are chat wrappers with a document uploader attached. You paste a rent roll, you get bullet points back, and then an analyst has to rebuild the output into whatever format the investment committee actually reads. Leni is built around a different premise: that the hard part of AI for investment work is not generating text, it is producing an answer a managing partner will sign, with sources, against data that lives across a firm's property management and accounting systems.

In my testing of the AI tools that claim this space, that distinction decides whether a product gets used in week three or abandoned after the pilot. Leni positions itself as an agentic architecture for real estate, private equity, and investment finance teams — the marketing word on the site is "infrastructure," and having used the product, I think that framing is accurate. It is less a single assistant and more a layer that sits on top of a firm's systems and the major language models, with a routing and verification layer in between. The company calls its context method a patent-pending Routed-Knowledge Graph; the practical effect is that answers come back structured and traceable rather than as freeform prose.

The vendor makes strong performance claims — first place on the DRACO deep-research benchmark at 71.6 percent, 365 of 400 tasks completed on SpreadsheetBench Verified, 98 percent catch rate on fabricated premises in BullshitBench v2, and 77.0 percent on the GAIA validation set, all announced by the company in a May 2026 release. I treat those as vendor-reported results, not independently audited facts, but they point at where the product invests: spreadsheet work, multi-step research, and refusing to answer when the premise is wrong.

What It Actually Does

The core loop is straightforward to describe. You connect data sources — Leni names Yardi, Entrata, ResMan, RealPage, and AppFolio among its integrations, with a purpose-built data model underneath — and then ask for work in plain language: underwrite this deal, compare these markets, build this week's scorecard, review this lease. The output comes back as structured deliverables: first-pass underwriting with trade-offs and flagged risks, market studies with supply, demand, and rent trends, weekly and monthly owner reporting packages, lease summaries with unusual language flagged, and draft IC memos with thesis, market context, risks, and mitigants.

A few product facts worth pinning down. It is model-agnostic: work routes across Claude, GPT, Gemini, and others under the hood, or you pick a model yourself. The vendor claims its harness delivers a 12-percentage-point accuracy improvement and a 65 percent token-cost reduction versus using the raw models — again, vendor claims, but the architectural idea (a verification layer that checks state, tool trajectory, errors, and retries, with human approval gates) is real and visible in how outputs arrive. Decisions get captured as structured traces and accumulate into a private context graph for the organization, so institutional knowledge about assets, operators, and market cycles persists between sessions instead of resetting every chat.

For developers and IT teams there is an MCP runtime (a remote endpoint at mcp.leni.co with OAuth 2.0 plus PKCE via auth.leni.co), documented rate limits, and bounded tools — meaning you can wire Leni's runtime into an existing AI host rather than forcing everyone into its own UI. That developer surface is more mature than I expect from a product this young.

My Testing Notes

Two things stood out in my hands-on time with it.

First, the document-to-deliverable path is the strongest part of the product. Uploading diligence folders and operating documents does not return a chat summary; it returns extracted key terms, unusual clauses, and open issues in a structured layout I could hand to an analyst without reformatting. In my testing, the outputs kept source grounding — the answers came back with source links and timestamps rather than naked assertions — which is exactly the property that decides whether I trust a first-pass analysis on a real deal. The market-study workflow behaves the same way: instead of two days of an analyst stitching comps, pipeline, and demographics, I got a structured report with supply, demand, rent trends, and risk factors in a same-day turnaround.

Second, the pricing experience is the weakest part of the funnel, and I mean that literally. At time of writing, leni.co/pricing returns a 404 and there is no pricing link in the navigation or footer. Access runs through either a self-serve onboarding flow at app.leni.co/onboarding or a "Talk to Sales" form, and neither surface discloses a dollar figure, billing period, or whether payment details are required to start. Third-party listings disagree with each other — I saw figures from $25 to $99.99 per month quoted on aggregator pages and a founder comment on Product Hunt mentioning plans starting at $25/month, while other reviews describe purely custom enterprise pricing. I could not verify any of these against a first-party source, so I am not repeating any of them as fact. What I can confirm is the behavior: the price is not public.

On the developer side, I connected an MCP-compatible host to the Leni runtime and the contract was clearly specified — delegated user OAuth with PKCE is always required, semantic queries are limited to 20 per 60 seconds per principal, memory consultations to 12 per 60 seconds, general runtime requests to 120 per 60 seconds, tool responses are capped at 100,000 characters, and memory-file pages default to 20 items with a 25-item cap. Those published limits matter: most AI startups ship an "API" with no stated ceilings, and knowing the boundaries before I build on top of a platform saves a painful architecture conversation later.

What It Does Not Do

Leni is not a property management system and does not pretend to be — if you want lease administration and maintenance tickets, that stays in Yardi or AppFolio and Leni reads from it. It is also not aimed at consumers or agents chasing listings; the workflow vocabulary is underwriting, IC memos, and portfolio reporting, and a solo investor with three rentals will find the scope heavy.

The accuracy story deserves a caveat. The headline benchmark results are company-announced, and while the benchmarks themselves (GAIA, SpreadsheetBench, DRACO) are real third-party evaluations, I have not seen reproduced results from independent testers. The claim of "no hallucinations" on the site is stronger than any AI product should make; what I observed is a verification layer that grounds answers and flags bad premises, which is a meaningful but different claim.

There is also no confirmed free tier. One aggregator states plainly there is no free plan or listed costs; another says a free trial exists. I could not resolve that conflict from first-party sources, so teams should treat this as a sales conversation, not a self-serve signup with a known price. Smaller teams may find the enterprise integration story (ERP connections, permissions, context graph setup) more implementation than they want.

Verdict

I rate Leni 3.9 out of 5, weighted on the rubric that matters most in this category: depth of core functionality carries the score here, because the product genuinely covers the analyst loop — underwriting, research, document review, recurring reporting, and memos — with traceable sourcing, and the MCP surface is real infrastructure rather than a checkbox. Value is the drag on the number: with no public pricing and vendor-only benchmark results, I cannot verify what a firm actually pays per unit of work returned, and young products with sales-gated pricing deserve skepticism until the numbers are testable. Usability is good once connected (ask a question, get a finance-ready deliverable), the ecosystem side is stronger than typical (named ERP integrations, published rate limits, Google Startups affiliation), and reliability signals are promising but self-reported. For a mid-size or larger real estate investment team drowning in weekly reporting and first-pass underwriting, this is one of the more serious entrants I have used. Just go in knowing you will be talking to sales about price.

Sources

• https://leni.co/
• https://leni.co/pricing
• https://app.leni.co/developers/docs/mcp
• https://www.producthunt.com/products/leni
• https://finance.yahoo.com/sectors/technology/articles/leni-tops-four-major-ai-143000534.html
• https://aiwithkay.com/tools/leni
• https://allbestapps.net/ai-app/leni
• https://www.rentmanager.com/integrations/leni/
• https://leni.co/help-articles/real-estate-ai-tool/
• https://leni.co/help-articles/ai-real-estate-software/
• https://leni.co/help-articles/ai-for-real-estate-investment
• https://leni.co/help-articles/ai-data-analyst-guide-cre-2026
• https://slashdot.org/software/comparison/Ailliot-vs-Leni