ResponseCX API

The API behind AI agents that resolve customer work — resources, auth, and where to start.

ResponseCX API

ResponseCX agents resolve customer work end to end: they look up the order, apply your policy, take the action, and close the ticket. This is the API that configures and observes them.

Agents are not chatbots. A chatbot returns text; a ResponseCX agent issues the refund. That difference is why the resources below are mostly about constraints — what an agent may do, and when it must defer.

Base URL and auth

https://api.stateset.com/v1

curl https://api.stateset.com/v1/agents \
  -H "Authorization: Bearer $STATESET_API_KEY"

Key prefix	For
sk_test_	Sandbox development
sk_live_	Production

Create keys in Dashboard → Settings → API Keys. They’re shown once.

The resources

Configuring behaviour — start here, because these decide what the agent is allowed to do:

Resource	Purpose
Agents	The agents themselves
Settings	Model, skip rules, escalation, human takeover
Rules	Named policy constraints
Functions · Function Parameters	Tools an agent can call
Attributes	Agent behaviour and knowledge settings

Conversation:

Resource	Purpose
Messages	Individual turns
Channel Threads	Threads grouping messages
Sessions	Conversation sessions
Responses	What an agent produced

Knowledge and memory:

Resource	Purpose
Embeddings	Vector representations for retrieval
Memories	Persistent per-customer context
Examples	Past interactions used as templates

Operations:

Resource	Purpose
Schedules	Scheduled agent work
Users	Operators and customers

Where to start

Set up ResponseCX

Connect integrations, configure an agent, first workflow.

Settings reference

The 22 fields that decide agent behaviour.

Public API (v1)

Build and tune agents programmatically.

MCP servers

Drive agent-building from Claude.

What sits underneath

The API is the surface. The parts that make an agent trustworthy live below it:

Layer	What it contributes
Workflow Studio	A 15-phase pipeline per brand, with a review gate
Temporal engine	Durable, replayable execution — an escalation can wait days
Decision gate	Per-action authorisation with cited proofs
Sync	186 integrations the agent can reach

Generative and deterministic operations are blended deliberately: the model reads the messy request, and deterministic workflows execute the consequential parts. Every action is traceable.

Related

• ResponseCX Platform — architecture and the outcome model
• Workflows — what teams automate first
• Authentication