Agent API / Private access

Use the engine.
Keep your infrastructure.

Add a production mortgage-engagement agent to your existing stack. You manage the channel, routing and CRM integration. SeedPath manages the conversation state, specialist model, tool decisions and structured outcomes.

Request API access Read the quickstart

Sandbox credits included

Reference / v1Stateful agent runtime

Customer-managed

SMS / WhatsApp

Routing

CRM sync

↓

SeedPath Agent APIAuthenticated

Thread state

Specialist model

Tool decisions

Validation

↓

Reply + structured outcome

thread th_91b8
intent book_appointment
action appointment.propose
validation passed

01 / Integration boundary

A conversational agent behind a REST API.

Send the agent a customer message and receive a reply together with any relevant appointment state and recommended follow-up. SeedPath retains the thread state between calls, so your application does not need to reconstruct or resend the conversation.

Designed to fit existing systems

Bring the channel.
We provide the agent.

SeedPath Agent API is designed for teams that already operate their own customer infrastructure: SMS or WhatsApp delivery, routing, CRM synchronisation and downstream workflows.

You integrate at a deliberately narrow boundary. The public API exposes the operations needed to run the conversation—not SeedPath's prompts, internal traces, graph or model configuration.

You manage

Your operational surface

Message delivery and receipt
Channel consent and opt-outs
CRM synchronisation
Customer routing
Downstream compliance records

SeedPath manages

The agent runtime

Persistent conversation state
Specialist agent behaviour
Appointment workflow state
Message and action idempotency
Response validation

02 / Quickstart

One thread. One message endpoint.

Create a thread once, then send each new customer message to that thread. The API credential establishes the customer boundary; every thread is isolated at the data layer.

01 / Create threadPOST

/v1/threadsrequest

{
  "provider_user_id": "usr_83921"
}

201 createdresponse

{
  "thread_id": "01JZ7YQ5H7Y1N8",
  "provider_user_id": "usr_83921",
  "created_at": "2026-07-21T09:58:12Z"
}

02 / Send messagePOST

/v1/threads/{thread_id}/messagesrequest

Authorization: Bearer <api_key>
Idempotency-Key: msg_18471

{
  "thread_id": "01JZ7YQ5H7Y1N8",
  "user_message": "Tomorrow morning would be ideal."
}

03 / Receive outcome200

structured responseapplication/json

{
  "provider_user_id": "usr_83921",
  "thread_id": "01JZ7YQ5H7Y1N8",
  "reply": {
    "text": "Would 10:00am tomorrow work?"
  },
  "appointment": {
    "status": "proposed",
    "datetime": "2026-07-22T10:00:00Z"
  },
  "follow_up": {
    "recommended_at": "2026-07-22T12:00:00Z",
    "message": "Would you still like me to arrange that appointment?"
  }
}

03 / Response contract

Minimal on the wire. Complete enough to act on.

A message call returns the content required by your channel and downstream workflow. Conversation history and internal execution traces remain behind the API boundary.

Reply, appointment state and recommended follow-up.

The response is deliberately structured around what your application needs to do next. When no appointment or follow-up exists, the relevant object is returned as null.

reply.text — the agent response for your customer channel
appointment — proposed, confirmed, cancelled or rescheduled state
follow_up — a recommended time and message for your own scheduler
processed_at — the UTC processing timestamp

response variantsJSON

// No appointment action
{
  "reply": {
    "text": "How can I help with your mortgage enquiry?"
  },
  "appointment": null,
  "follow_up": null,
  "processed_at": "2026-07-21T10:00:00Z"
}

// Appointment confirmed
{
  "appointment": {
    "status": "confirmed",
    "datetime": "2026-07-22T10:00:00Z"
  }
}

04 / Agent configuration

Your business context. Our operating controls.

Configure how the agent represents your organisation without taking responsibility for the underlying prompt, orchestration or validation system.

Structured configuration—not unrestricted prompt access.

Working hours, business context, available services, referral relationships and tone can be managed through a validated, versioned API.

Customer configuration can influence expression and business context. It cannot override SeedPath safety controls, action constraints, system-level instructions or model configuration.

Configurable

Working hours
Business summary
Tone profile
Services offered
Referral partners

SeedPath-controlled

Safety boundaries
System prompts
Validation logic
Orchestration graph
Model configuration

PATCH /v1/configIf-Match: "cfg_12"

{
  "timezone": "Europe/London",
  "working_hours": {
    "monday": [
      {
        "opens_at": "09:00",
        "closes_at": "18:00"
      }
    ]
  },
  "business": {
    "name": "Example Mortgages",
    "summary": "Independent UK mortgage brokerage."
  },
  "tone": {
    "profile": "warm_direct"
  },
  "services": [
    "residential_mortgage",
    "protection"
  ]
}

200 OKconfiguration versioned

{
  "configuration_version": 13,
  "status": "active",
  "updated_at": "2026-07-21T10:14:22Z"
}

05 / State and deletion

State is managed. The machinery remains abstracted.

Each thread retains only the conversation and workflow state required by the agent. Customers create, use and delete threads through the public API; internal prompts, traces and routing decisions are not exposed.

Active thread

Conversation state is retained for up to 400 days to support long-running mortgage journeys.

DELETE request

Message content and conversational state are removed immediately from the active production datastore.

Backups

Deleted records may remain temporarily in encrypted nightly backups until those backups expire under the backup-retention schedule.

Reporting layer

Limited operational metadata is retained without message content, contact details or the customer-supplied user identifier.

06 / Reliability contract

Predictable behaviour at the integration boundary.

The API is versioned and designed for safe retries. Errors are structured so your application can distinguish authentication, validation, rate-limit and service conditions.

01 / Authentication

Customer-scoped credentials

API credentials establish the customer boundary. A credential cannot access threads belonging to another customer.

02 / Idempotency

Safe retries

Repeating a request with the same key returns the original result without creating a second thread, reply or appointment action.

03 / Versioning

Stable contracts

Public endpoints are versioned under /v1. Breaking changes are introduced through a new API version.

04 / Request IDs

Traceable without traces

Every response includes a request identifier for support and operational investigation without exposing internal execution.

Errors your application can act on.

Successful message calls return a structured agent response. Invalid requests, access failures and temporary service conditions use stable machine-readable codes.

error responseapplication/json

{
  "error": {
    "code": "thread_not_found",
    "message": "No thread was found for the supplied identifier.",
    "request_id": "req_01JZ7Z2EXAMPLE"
  }
}

07 / Pricing

Per call, not per seat.

Start in a sandbox, integrate against representative traffic and move to usage billing only when you are ready for production.

No platform subscription

No minimum commitment during evaluation.

Sandbox access includes free calls for integration and representative message testing. Production usage is charged per agent message call. A payment method is only required when the free allowance has been used or a workspace moves into production.

Request sandbox access

Commercial model

Free sandbox allowance
No per-seat charge
No platform fee to begin testing
No minimum evaluation commitment
Usage billing when production begins

08 / Who it is for

Conversational AI as a component.

For organisations that want specialist agent capability inside their existing product and operating environment—not another platform to migrate onto.

Brokers and networks

Technology-led operators

Teams with their own channel, CRM and engineering capability that want conversational AI as an infrastructure component.

Software and agencies

Integrate rather than rebuild

Organisations that would rather integrate a specialised mortgage agent than build, evaluate and maintain one internally.

Distribution partners

Add the model layer

Partners that own customer acquisition or distribution and want mortgage-conversation capability without becoming a model-development company.

09 / Choose the route

Agent API or managed system?

Both routes use the same underlying agent and model programme. The difference is how much of the surrounding operation you want to own.

Layer	Agent API	Managed system
Channels	Customer-managed	SeedPath-managed
Message routing	Customer-managed	SeedPath-managed
CRM integration	Customer-managed	SeedPath-managed
Conversation state	SeedPath	SeedPath
Specialist agent	SeedPath	SeedPath
Model evaluation	SeedPath	SeedPath
Operational monitoring	Customer owns integration	SeedPath manages end-to-end
Optimisation	Agent and model level	Agent, workflow and channel level

10 / Getting started

From credential to production.

Private access begins with a scoped sandbox environment and representative message testing.

01 / Request access

Scoped sandbox credentials

Receive an API credential and free test-call allowance.

02 / Integrate

Create and message threads

Connect your channel and test representative conversation sequences.

03 / Validate

Test the boundaries

Confirm routing, state handling, idempotency and downstream actions.

04 / Production

Activate usage billing

Issue production credentials and begin processing live message calls.

Bring your infrastructure.
Use our agent.

Tell us what you are building and how you intend to integrate it.

Request API access →Trust & Compliance