Ask twelve building-software vendors whether their platforms are open and you will get twelve yeses. This report card asks a narrower question: what does the public evidence support?
Below we score twelve major platforms — BMS incumbents, middleware, property management systems, and CRE data platforms — on five data-openness criteria, using only facts verified from public materials in our June 2026 research pass. No briefings, no demos, no inference. Where the public record gave us nothing, we say so rather than guess.
This is v1 of a living document. Every vendor on this list has a right of reply, detailed at the end.
The rubric, stated up front
Five criteria, each scored 0-2, for a total out of 10:
- Public API documentation exists. Can a developer find documented APIs without talking to sales?
- Third-party access without mandatory partnership or certification. Can a third party reach data without joining a partner program?
- Standard data export documented. Are CSV/trend exports a documented, supported capability?
- Open ontology support stated. Does the platform publicly state support for Project Haystack, Brick Schema, or ASHRAE 223P?
- Agent/developer surface. Is there an MCP server, public SDK, or genuinely open marketplace?
Scoring: 2 = open and publicly documented. 1 = exists, but gated or partial. 0 = closed or absent per public evidence. ID = insufficient data — no reliable public evidence either way; awards no points and marks the row provisional (†). Letter grades: A (9-10), B (7-8), C (5-6), D (3-4), F (0-2). Four or more ID cells: no letter — recorded as ID.
One framing note: this rubric measures openness as a third-party developer experiences it. Capabilities may exist behind login walls or sales conversations; for an openness audit, that distinction is the point.
The scores
| Platform | C1 API docs | C2 Access w/o partnership | C3 Export | C4 Ontology | C5 Agent surface | Score | Grade |
|---|---|---|---|---|---|---|---|
| Tridium Niagara | 1 | 1 | 2 | ID | 1 | 5/10 | C† |
| JCI OpenBlue | 1 | 0 | ID | ID | 0 | 1/10 | F† |
| Siemens Building X | 1 | 0 | ID | ID | 0 | 1/10 | F† |
| Yardi | 1 | 0 | ID | ID | 0 | 1/10 | F† |
| Honeywell Forge | ID | 0 | ID | ID | 0 | 0/10 | F† |
| Schneider EcoStruxure | ID | 0 | ID | ID | 0 | 0/10 | F† |
| CoStar | 0 | 0 | ID | ID | 0 | 0/10 | F† |
| Altus ARGUS Assist | 0 | 0 | ID | ID | 0 | 0/10 | F† |
| Trane (+BrainBox) | ID | ID | ID | ID | 0 | 0/10 | ID |
| RealPage | ID | 0 | ID | ID | ID | 0/10 | ID |
| MRI | ID | 0 | ID | ID | ID | 0/10 | ID |
| Cherre | ID | ID | ID | ID | 0 | 0/10 | ID |
Scoring notes — every scored cell, traced
Tridium Niagara (5/10, C†). C1=1: Niagara Data Service APIs are publicly documented in vendor Q&A materials, though not as full self-serve reference docs [source: Tridium Niagara Data Service APIs Q&A, 2024]. C2=1: access is gated by paid per-site module licensing — a commercial gate, not a certification program [source: Tridium licensing materials; practitioner verification, 2026-06]. C3=2: scheduled CSV/trend export from JACE supervisors via FTP/SFTP/email is standard, universal practice [source: practitioner verification, 2026-06]. C4=ID: the practitioner ecosystem routinely pairs Niagara with Haystack tagging, but we did not verify a vendor statement of open-ontology support. C5=1: a real third-party module marketplace exists (e.g., ossEasyAPI on Niagara Marketplace); no MCP server [source: Niagara Marketplace listings; MCP registry census, 2026-06].
JCI OpenBlue (1/10, F†). C1=1: a cloud API exists; access is onboarding-mediated [source: company platform documentation, 2026]. C2=0: vendor-cloud onboarding required; supervisory access on Metasys is gated by Site Director licensing [source: company documentation; practitioner verification, 2026-06]. C5=0: no MCP server or open agent surface found in a census of 10,000+ active MCP servers [source: MCP registry census, 2026-06]. C3, C4: ID.
Siemens Building X (1/10, F†). C1=1: a partially open, ecosystem-preferred API surface ("open-ish") [source: company documentation review, 2026-06]. C2=0: vendor-cloud onboarding required [source: same review]. C5=0: no MCP or open agent surface found [source: MCP registry census, 2026-06]. C3, C4: ID.
Yardi (1/10, F†). C1=1: SOAP/REST interfaces exist [source: practitioner verification, 2026-06]. C2=0: access gated by a partnership program and per-client keys; independently characterized as an API walled garden [source: practitioner verification; industry review, 2026]. C5=0: the Virtuoso agent platform is closed, with no public developer surface [source: vendor materials review, 2025-2026]. C3, C4: ID.
Honeywell Forge (0/10, F†). C2=0: vendor-cloud onboarding required [source: company documentation review, 2026-06]. C5=0: AI capability delivered via the 2024 Gemini partnership inside the platform; no open agent or developer surface found [source: press, 2024; MCP registry census, 2026-06]. C1, C3, C4: ID.
Schneider EcoStruxure (0/10, F†). C2=0: ecosystem-preferred; supervisory integration is certified-integrator-only [source: company documentation; practitioner verification, 2026-06]. C5=0: no MCP or open agent surface found [source: MCP registry census, 2026-06]. C1, C3, C4: ID.
CoStar (0/10, F†). C1=0: no public API — confirmed [source: vendor surface review, 2025-2026]. C2=0 and C5=0 follow directly: with no API there is no third-party access path and no developer surface [same source]. C3, C4: ID.
Altus ARGUS Assist (0/10, F†). C1=0: no API — confirmed; the product also does not touch BMS/IoT data [source: vendor materials review, 2025-2026]. C2=0: no API, no third-party access path [same]. C5=0: an AI assistant without a developer surface [same]. C3, C4: ID.
Trane +BrainBox (ID). C5=0: BrainBox AI — running across 14,000+ buildings — was acquired by Trane in a deal closed January 3, 2025, and now operates inside Trane's ecosystem; no open agent surface documented [source: Trane investor release, 2025-01]. C1-C4: ID. Too little public evidence to grade.
RealPage (ID). C2=0: characterized as an API walled garden [source: industry review, 2026]. Worth noting: the DOJ proposed consent judgment (2025-11-24) restricts RealPage's use of nonpublic competitor data in pricing algorithms — it does not mandate API openness [source: DOJ filing and legal analyses, 2025-2026]. C1, C3, C4, C5: ID.
MRI (ID). C2=0: named among PMS API walled gardens [source: industry review, 2026]. Everything else: ID. Per protocol, we do not guess.
Cherre (ID). C5=0: the Agent.STUDIO agent offering is closed [source: vendor materials review, 2025-2026]. Cherre's core data-integration APIs were not assessed in this pass — C1-C4: ID. We expect this row to change in v2 and invite the correction.
What the table says
Eight points out of 120 possible. Twelve platforms, ten points each; combined industry score: 8. One C, seven F's, and four rows too thinly documented in public to grade at all.
The top grade belongs to twenty-year-old middleware. Tridium Niagara earns the only C the boring way: documented (if licensed) APIs, universal trend export, a real third-party marketplace. No AI-era platform on this list came close.
Criterion 4 produced zero points across all twelve. Not one platform, in the evidence we could find, publicly states support for Haystack, Brick, or 223P. Some of that reflects the standards themselves — 223P remains proposed, not published. But the asymmetry stands: every vendor on this list markets AI; none of them markets open semantics.
Criterion 5 is nearly as empty. In an ecosystem of more than 10,000 MCP servers, the only building-domain server we found is a community BACnet prototype — built by no vendor on this list.
Thirty-three of sixty cells are ID. More than half the rubric could not be scored from public evidence. That is not a weakness of this report card; it is the report card: when the public record on data access is this thin, the opacity is the finding.
Right of reply
Built from public evidence, this card will be wrong in both directions. Vendors may submit corrections, with supporting documentation, to hello@ai-smart-buildings.com. Every substantiated correction will be incorporated into v2, with every regrade published in a changelog. ID cells are an open invitation: show us the documentation and the score changes.
Methodology and limits
Evidence: a June 2026 research pass across 24+ primary and secondary sources — vendor documentation, regulator and court filings, press, GitHub, and an MCP registry census. Scope: the named platform surface, not a vendor's full portfolio. We scored what a competent third-party developer could discover; a capability that cannot be found functions, for that developer, as closed. Grades carrying any ID cell are provisional (†). v2 will rescore all twelve, plus reader-nominated platforms, after the right-of-reply window closes.
Research compiled by the AISB agent fleet from primary sources; every claim verified against the public record. Cost figures are labeled industry estimates. Full source list available on request — hello@ai-smart-buildings.com.