Can Claude Read ARGUS? What Altus Has Shipped, and What It Hasn’t

Altus Group’s AI move so far happens inside ARGUS. On April 14, 2026, at its Altus Connect conference in Palos Verdes, California, Altus unveiled ARGUS Assist, which it describes as the first AI experience layer on its ARGUS Intelligence Platform: a conversational interface embedded across the platform, where users ask questions or request analyses in plain language and specialized AI agents carry out tasks across ARGUS. Altus says the feature draws on the structured financial models, valuation frameworks and market data already in ARGUS, was designed with strict data isolation, and keeps all AI processing inside the secure ARGUS environment. Its chief product and technology officer framed the goal as moving ARGUS from a system of record for financial modeling toward a platform for performance analysis. Those are company claims. The release does not mention Claude, Anthropic or the Model Context Protocol, and it describes no way for an outside assistant to reach ARGUS data.

Anthropic’s side of the record is an absence. Its connector directory, which listed more than 800 entries when PSV checked it on September 24, 2026, had no connector from ARGUS or Altus Group; the real estate data connectors it did list were Yardi Virtuoso, Yardi Matrix, AppFolio Realm-X and MSCI. That check covers Claude only; PSV did not check other assistants’ app directories. What Claude can do is read files. Anthropic’s help article on uploads, dated July 23, 2026, lists PDF, DOCX, CSV and XLSX among supported document types, with XLSX requiring code execution and file creation to be enabled, and allows up to 20 files of up to 500MB each per chat, with PDFs capped at 1,000 pages. For PDFs, Claude analyzes charts and other visual elements only in files of 100 pages or fewer, and processes text only in PDFs of 101 to 1,000 pages. So the practical route to Claude runs through reports your team already produces as Excel or PDF, not through ARGUS itself. Altus also publishes an ARGUS API fair usage policy, so any firm-built integration is a question for the firm’s Altus agreement, not something to wire up from a staff login.

The split matters because the two routes answer different questions. ARGUS Assist, on Altus’s description, works where the model lives: it can query and analyze the cash flows, valuations and market data inside ARGUS without anything leaving that environment, which suits questions about one model. Claude works where the rest of the evidence lives: the executed leases and amendments, the rent roll from the property management system, the lender’s term sheet, the offering memorandum. A familiar underwriting error sits between those two worlds: a rent step modeled in the wrong month, a recovery structure entered as full pass-through when the lease sets a base-year stop, a renewal option modeled as exercised when the tenant has said nothing. Neither assistant closes that gap alone. Altus’s release does not describe ARGUS Assist reading lease documents, and Claude does not see the ARGUS model unless someone exports it.

That leaves two gates before any ARGUS report goes into Claude. The first is contractual. PSV reviewed Altus’s published legal page on September 24, 2026 and found policies on privacy, client data and corporate conduct, and an ARGUS API fair usage policy on its ARGUS Enterprise downloads page, but no published terms that address AI tools, and a web search turned up none either, so the governing words sit in each firm’s own Altus agreement, and that is the document to read before exporting anything to an AI tool. The second is the data inside the model. An ARGUS file often carries tenant names, lease economics and, on an acquisition, a seller’s rent roll received under a confidentiality agreement, and each of those sources has its own terms about where it may go. Clear both, use a Claude plan under Anthropic’s commercial terms, and the export route is straightforward. Skip them, and a convenient reconciliation can become a breach of a license or a confidentiality agreement. This is a reading of published documents as of September 24, 2026, not legal advice.

The job is a lease-to-model reconciliation, run before an investment committee memo or a quarterly valuation. Inputs: the ARGUS cash flow and tenant-level rent schedule reports the team already produces as Excel or PDF, plus the executed leases, amendments and lease abstracts for the same property. Claude extracts, clause by clause, the terms that drive the model: base rent and every scheduled step with its effective date, free rent, the recovery method with any base year, stop, cap or exclusion, renewal and termination options, and expiration. The comparison itself runs in code rather than in the chat: a short script lines up each tenant’s lease terms against the exported ARGUS rows and flags every difference in amount or date, so the arithmetic is reproducible and a reviewer can re-run it. Output: an exceptions register with one row per mismatch, citing the lease page and the report row, and the dollar effect computed from the export rather than estimated in prose.

The reviewer is the model owner, the analyst or asset manager accountable for that ARGUS file, and the review runs from the register, not from Claude’s summary. The approval gate is simple and absolute: every change is made in ARGUS by that person, who records what changed and why, and the next export goes through the same comparison to confirm the register has cleared. Nothing flows back into ARGUS automatically, and nothing Claude drafts reaches a committee or a lender until the model owner signs the reconciled file. Start with one property whose model the team already trusts, so the first run tests the workflow rather than the deal, and keep the register as part of the deal record. PSV has not run this workflow against ARGUS Assist or measured either tool’s accuracy on ARGUS reports; it is the order of operations PSV would use, not a performance claim.