HabeoFeaturesCopilot

Feature · AI Copilot · updated May 19, 2026

The Copilot built on the institutional ledger.

Habeo Copilot is the AI assistant for university IT — not a chat layer bolted onto a generic CMDB. It reads from your live MDM and Workday, speaks GASB 34/35 and Uniform Guidance, proposes writes with a confirmation flow, and signs every step into an immutable audit log.

12
Grounded tools
3
Surfaces · App · Chat · Query bar
0
Agents to install
0-retention
Enterprise tier
Why a different AI for higher ed

Two ways to put AI inside ITAM.

General-purpose ITAM AI

Trained on the generic IT asset universe.

Doesn't know what a college unit is. Treats a grant-funded NSF microscope the same as a printer. Asks you to teach it your data model.

Habeo Copilot

Trained on the higher-ed institutional ledger.

Speaks GASB 34/35, Uniform Guidance §200.313, FERPA-aware data handling, InCommon, and the org-unit reality of an R1 system. Out of the box.

Three surfaces, one Copilot

Where the AI shows up.

Same retrieval, same tools, same audit trail. Three places to call on it.

In-product Copilot

/assistant · floating widget on every page

Full-page chat plus a docked widget on every Habeo page. Answers stream in token-by-token; threads persist; every tool call is shown in the transcript with the rows it touched. "Where is ORG-MBP-22841 and who has it?" comes back grounded against the live ledger row, not a model recollection.

Slack & Microsoft Teams

OAuth · signed Events · Bot Framework v4

Habeo Copilot meets your team where work already happens. "When is my MacBook up for refresh?" "Pull the surplus list for Riverpoint Physics." Approvals get signed off in-line with adaptive cards in Teams and rich blocks in Slack — no context switch into another tool.

Natural-language query bar

/assets · /people · /saas · /loaners

Embedded on every list page. Type "all Dell laptops in Engineering due for refresh in FY26" and the query is translated to a structured filter spec — not free-form generation. You see the spec, you can edit the chips, and the model never invents columns that aren't in the schema.

What it looks like

One question, three grounded answers, an audited write.

How it stays honest

Answers come from the ledger, not the model.

Habeo Copilot never speaks about your assets without first retrieving them. Every claim cites the row that backs it.

Context pack on every turn.

Each chat turn ships a structured grounding block to the model: caller identity, role, college-unit scope, top-K asset matches via embeddings, top-K people, and the last 5 audit events touched by the caller. Hallucinations starve.

context-pack.ts

Embeddings, scoped to your org.

OpenAI text-embedding-3-small, indexed inside the tenant boundary. The Copilot never retrieves across organizations — and per-row tenancy is enforced server-side, not in the prompt.

embeddings.ts · pgvector

College-unit RBAC for the AI itself.

A Westmark asset manager asking about "all laptops" only sees Westmark laptops — including from the Copilot. The same college-unit scope that gates the UI is threaded through every tool call.

ctx.collegeUnitIds
Reads · and proposed writes

An assistant that does the thing, not just talks about it.

Twelve tools today. Eight read-only, four write-proposing — each write blocked behind an explicit human confirmation flow before it touches the ledger.

Read tools
No confirmation needed
  • asset_findSearch by tag, owner, dept, status, refresh FY.
  • asset_get_by_idPull a single asset with relationships.
  • asset_list_by_assigneeEverything held by a named person.
  • person_findLook up by name, email, or supervisor.
  • recent_activity_for_assetLast 10 audit events for a row.
  • assets_on_grantProperty under a federal award.
  • federal_property_inventoryUniform Guidance §200.313 export.
  • surplus_disposalsRecently disposed assets, with reason.
Write proposals
Confirmation flow required · written to audit log
  • propose_asset_assign_personAssign an asset to a person.
  • propose_asset_returnMark an asset returned to the holding pool.
  • propose_asset_transition_statusMove status — deployed, refresh, repair, retired.
  • propose_ticket_linkBind an asset to a ServiceNow CHG/INC.
For your CISO

AI your security review will sign off on.

Higher-ed IT lives or dies by the audit trail. Habeo Copilot was built to be auditable from day one — not retrofitted with logging when the first SOC 2 came around.

Immutable audit log.

Every tool call, every prompt, every confirmed write — written to the same append-only ledger that backs the rest of Habeo. Auditors get one place to look.

Zero-retention enterprise tier.

Anthropic enterprise-tier inference via Vercel AI Gateway. Your prompts are not retained, not used for training, not seen by Anthropic personnel.

Per-org budget caps.

Configure a monthly dollar cap from /settings/ai. The gateway pre-flights every call against remaining budget and returns a BudgetExceededError before a token leaves your org.

College-unit RBAC.

The same scope that gates the UI gates the Copilot. A Westmark asset manager can’t see Liberal Arts inventory through the AI any more than they can through /assets.

FERPA-aware data handling.

Student-PII fields never enter the model context unless the caller explicitly retrieves them. The Copilot is configured to refuse aggregate queries that would re-identify under FERPA.

Off-list-model deny.

Only the model IDs Habeo has procurement-approved can be called. New providers go through HECVAT review, not a config flag.

The model stack, in the open

No magic. One gateway, four approved models.

Every Copilot call routes through Vercel's AI Gateway. Habeo never talks to a model provider directly — so every call gets the same auth surface, observability, and budget enforcement.

RoleModelProviderWhen it's used
Default chatClaude Sonnet 4.6AnthropicDay-to-day Q&A, tool orchestration, drafting writes.
Hard reasoningClaude Opus 4.7AnthropicLifecycle disposition under grant constraints; multi-step compliance reasoning.
Cheap readsClaude Haiku 4.5AnthropicTool argument parsing, single-row lookups, NL-query translation.
Embeddingstext-embedding-3-smallOpenAIAsset and people retrieval for the context pack.

Routing · fallbacks · observability via @ai-sdk/gateway

Habeo Copilot vs general-purpose ITAM AI

A side-by-side, on the points that decide procurement.

We compete most often against AI features in general-purpose ITAM tools like Lansweeper. Honest read on the differences:

What you're comparingHabeo CopilotHigher-ed-native AIGeneral-purpose ITAM AILansweeper AI · category
Grounding modelHigher-ed institutional ledger. MDM + Workday + grants + GASB classes, joined inside the tenant.Generic IT asset graph from agent-based discovery. Bring-your-own data model for higher-ed concepts.
Write actionsProposed writes for assign, return, status, ticket-link — confirmation required, audit-logged.Largely read-only AI summarization. Writes typically still happen in the legacy console.
Slack · Teams nativeOAuth Slack app + Teams adaptive cards ship in production. Approvals signed off in-line.Limited; partial via integrations or third-party connectors.
Higher-ed fluencyGASB 34/35, Uniform Guidance §200.313, FERPA-aware, InCommon, college units — out of the box.Not built for higher ed. No native model of grants, capital classes, or college unit hierarchies.
Audit postureEvery tool call + every confirmed write written to the same immutable audit log as the rest of the product.AI calls logged separately from the ITAM audit trail; correlation is left to the customer.
Data retentionZero-retention enterprise tier on Anthropic via Vercel AI Gateway. Off-list-model deny enforced.General-tier provider terms; customer-side review required.
Org-scope guardrailsCollege-unit RBAC threaded through every tool call. A unit-scoped admin only sees their unit.Org-wide retrieval by default; scope segmentation depends on the customer's configuration.

Drawn from public product pages and contracts our prospects have shared in side-by-side procurement reviews. Last reviewed May 19, 2026. See the full Compare hub.

Frequently asked

Copilot, answered.

The questions we hear in every procurement review. Answered briefly, with the technical detail your CISO will ask for next.

Does the Copilot ever make changes without my approval?
No. Reads return data directly; every write goes through a propose → confirm flow. Until a human clicks Confirm, the ledger is untouched. Confirmed writes are signed into the immutable audit log alongside the originating thread and the model that proposed them.
What does it cost to run?
All Copilot inference is billed inside your Habeo institutional contract — there is no per-token line item to manage. You set a monthly USD cap per organization from /settings/ai; the gateway pre-flights every call and returns a BudgetExceededError if the cap would be breached.
What models does it use?
Claude Sonnet 4.6 by default, Opus 4.7 for hard reasoning, Haiku 4.5 for cheap reads, and OpenAI text-embedding-3-small for retrieval. All routed through Vercel AI Gateway on the enterprise zero-retention tier. New models are added behind a HECVAT review and an off-list-model deny at the gateway.
How is this different from Lansweeper AI?
Lansweeper AI is a chat layer on top of a general-purpose IT asset graph built from agent-based discovery. Habeo Copilot is the AI for a system of record purpose-built for higher ed — it grounds in your live MDM and Workday data, speaks GASB 34/35 and Uniform Guidance §200.313 natively, proposes writes with confirmation, and threads college-unit RBAC through every tool call.
Is student data exposed to the model?
Student PII never enters the model context unless the caller has explicit permission to read it through the UI. The Copilot also refuses aggregate queries that would re-identify under FERPA. Every prompt and tool call lands in the immutable audit log for compliance review.
Can I disable it institution-wide?
Yes — org admins can disable the Copilot entirely from /settings/ai, restrict it to read-only mode (no propose_* tools), restrict surfaces (turn off Slack and Teams while keeping in-app), or apply a $0 monthly budget cap to soft-disable.
See it answer your real questions

Bring your messiest asset question.

On the demo call we'll connect your sandbox MDM and Workday tenant and let the Copilot answer the question your team currently spends an afternoon on. 30 minutes, founder-led, no SDR triage.