01
AI systems and agents
Agents and models that do the machine work inside limits a person sets.
In our work
- One agent per account reads the dossier every cycle, decides the next step and logs it.
- A router chooses the data source and the model for each job, with automatic fallbacks, so no single vendor is the product’s AI.
- Raven’s vision research, in development, is aimed at the judgment calls of the trades: what a failing weld or a tired bearing looks like.
Hyves · Raven · Hangar
02
Research and evidence engineering
Research built so every claim points at its source, and a gap stays a gap.
In our work
- Research on a new user’s own company runs 21 stages in 7 chapters, and evidence not anchored to that domain is thrown out.
- A 12-dimension rubric written for each product, every dimension scored 0, 1 or 2 with the evidence quoted; missing evidence counts as unknown, never zero.
- Every business case names its signal, source and date, and an empty case means nothing was found; it is never invented.
Hyves · Hangar
03
Data and integrations
Connecting the systems a team already runs, and keeping what they hold in one record.
In our work
- 35 one-click connectors, plus 1,000+ apps through Composio; OAuth tokens are held by Composio and never stored by Hyves.
- The same company or person is resolved across connected systems into one canonical record, in one account memory shared by every module.
- Raven’s research program includes a craft memory: the same one-memory architecture, pointed at the physical world.
Hyves · Raven
04
Human-in-command interaction design
Designing how a person commands a system: what it may do alone, what waits for them, and what it shows them first.
In our work
- Each Hyves account runs in a send mode its team chooses: approve every message, approve the first, or let it run inside the plan.
- Hyves does not dial; the rep gets a brief and six to eight questions aimed at what the business case does not know yet, and tasks from the call are proposals to confirm.
- Any change in a connected app waits for approval, and Hangar teaches builders to write down what an agent may do unattended before their first hire.
Hyves · Hangar
05
Field research
Going where the work happens, with the people who do it, before any product assumption is made.
In our work
- Field work with the people affected is the second of five stages every division walks; Raven and the foundation work in it today.
- Raven’s research runs on job sites with working tradespeople, mapping where minutes and knowledge actually leak.
- The foundation’s model works backwards from the work a place will be short of, and only then decides what to teach and what to reward.
Raven · Hyve Foundation
06
Privacy and security engineering
Isolation and data rules enforced in the system itself, and described no wider than they are.
In our work
- Tenant isolation is enforced in the database by row-level security, not in the interface alone, with role-based access control across every module.
- Hyves is hosted on Supabase in eu-west-1 and workspace data lives in the EU; some research and messaging providers sit outside it, under DPA safeguards.
- Hangar’s fourth level publishes the checks for builders: identity from the verified session, never the request body, and no service key reachable from the browser.
Hyves · Hangar