Agent access
Ten sources, and the answer stops arriving
An on-site assistant invents a compatibility answer. A buyer's procurement agent cannot reach your catalogue at all, because your ERP will not open to an external caller.
Enterspeed Core is an AI data orchestration platform that composes your product truth once, and Agent Context is the read-only surface an agent calls to reach it. Every response is shaped by the caller's identity, so a logged-in session sees entitled prices and a procurement agent from another organisation sees what you publish. Your commerce platform, your ERP and your PIM stay where they are.
92% to 44%
Reliability of a naive multi-source tool call, one source against ten
Our own measurement rather than a customer result, and we can document the method if a technical buyer asks for it. Reliability compounds downward with each source added, which is the argument for preparing the read rather than fanning out at request time.
How it works
Discover, query, fetch
Agent Context exposes a read-only interface with three operations, and the caller's identity is enforced at every one of them.
Discover
The agent asks what exists and receives a machine-readable statement of the catalogue, the markets you operate in, the actions allowed and the identity required. The schema is derived from the data, so it stays current as your catalogue changes.
Query
The agent asks for products by any combination of attributes and receives composed facts, with an explicit statement of what is unknown rather than a guess. The policy is enforced before the query runs, not by filtering afterwards.
Fetch
The agent retrieves a single document by identifier. Fields carry usage guidance, so the agent understands what it is reading and how to interpret a null value.
The same interface serves an on-site assistant grounding its answers in real compatibility data and an off-site procurement agent discovering your catalogue without touching your ERP.
Scope
What your systems keep
Agent Context is the surface an agent reads. It is not an agent and not a chatbot, and it does not decide what the agent does with what it receives.
Agent Context carries
- The composed read, in one call
- The entitlement check, before the query runs
- The usage guidance travelling with each field
- The access log of who fetched what
Your systems keep
- Your commerce engine: pricing calculation and payment
- Your ERP: inventory management and order fulfilment
- Your agent platform: the reasoning, the prompt and the next step
The agent calls Agent Context, receives structured truth and decides what to do with it. The systems holding the source data never open to the agent directly, and the agent never writes to anything.
