NATA Orchestration

NATA coordinates complex work across the Nata ecosystem. In NataPulse, orchestration connects the data layer to specialist reasoning without exposing the private control plane to the end user.

Orchestration responsibilities

For a supported task, the orchestration layer can:

interpret the user’s requested scope;
select the appropriate workflow;
retrieve relevant events, clusters, reports, and memory;
limit context to the workspace and task;
decompose work into specialist steps;
route steps to logical agents;
coordinate dependencies and hand-offs;
enforce permissions and research wording;
collect evidence and citations;
synthesize a product-safe result;
record applicable feedback or memory.

Logical agents versus providers

A logical agent is defined by its role and contract, not by one permanent model vendor. Provider and model selection can change independently from the product-facing role.

This separation allows NataPulse to improve or replace underlying models without changing the meaning of “Fundamental Analyst,” “Risk Reviewer,” or “Research Supervisor.”

Context control

An agent should receive only the context required for its task. A technical analyst does not need raw credentials. A product report does not need infrastructure traces. A public answer does not need private administrative scores.

Context bundles can include:

published event evidence;
source references;
relevant market context;
prior reports;
safe memory lessons;
the requested horizon and question;
explicit limitations.

Failure behavior

A failed specialist step should not silently become a successful conclusion. The workflow can retry according to policy, mark the step failed, continue only when safe, or fail the run with an honest status.

Product boundary

NATA’s internal coordination is broader than what the product exposes. NataPulse shows useful progress, sources, findings, confidence, and limitations while excluding internal secrets, cost data, raw traces, and non-public control-plane fields.