Types
Entities, relations, states, derived facts
tend is two halves of one platform. A standardized data layer your agents can actually read, plus an agent runtime that ships them on top to do the work nobody on your team wants to do. Replaces the fragile glue between your tools, not another one of them.
Define the customers, deals, and processes your team actually operates on. What an account is, what a renewal looks like, who owns what, how things relate. Engineers call this an ontology. It's the difference between AI that hallucinates and AI that gets the job done. Version it, diff it, migrate it without downtime. Every agent reads from here.
# schema / account.tend entity Account { id: uuid name: string tier: enum("smb", "mid", "ent") health: computed { risk_score() < 0.3 ? "green" : "yellow" } owned_by: Contact via "primary_owner" }
Pull from 180+ managed connectors or stream CDC from your warehouse. Deterministic plus probabilistic entity resolution merges duplicates across systems while keeping full provenance.
Attribute-level permissions, row-level policies, versioned decision logic. Every read and write, from humans or agents, passes through the same rules. Full audit trail by default.
# policy / agent_revenue.policy policy agent_revenue { can read Account, Deal, Contact can write Deal.notes, Contact.last_touch cannot read Account.pii where region == "EU" rate: 120 ops/min, $50/day }
Ship agents on top of the data layer to chase renewals, route tickets, reconcile invoices, and other busywork. They read your model as real context, propose actions, and respect the same rules as humans. Triggers fire on data changes; writes go back to your tools.
The platform ships as one product, not nine. These are the surface areas your team will touch.
Visual + code editor with diffs, migrations, and test runners.
180+ managed connectors, declarative SDK for the rest.
Rule + ML entity resolution with full provenance.
Temporal, columnar, and vector indexes in one engine.
ABAC, row-level rules, test harness, full audit trail.
GraphQL, SQL, and natural language queries, all governed.
Model-agnostic, typed context, bounded tools, replay-safe.
Event-driven, retries, SLAs, saga support, write-back.
Lineage, run history, token & cost reporting, alerts.