Agent quickstart

Use this guide when you want Codex, Claude, Cursor, or your own integration to work with an Audryo workspace. The API is designed for agents: it exposes the same reviewed lifecycle workflow as the app, with no browser automation needed.

What an agent can do

An agent can read product context, sync data, propose audiences, build Journeys, and draft localized emails. It does not get to silently activate a Journey, schedule a campaign, or send customer email. Those delivery actions stay explicit and reviewable.

1. Create a workspace API key

In Audryo, open Settings → Agent access and create a token with the smallest useful scope:

Copy the secret when it is shown. It is intentionally displayed once. Store it in your agent's encrypted environment, never in a prompt, source repository, or application log.

2. Give your agent this starting prompt

Replace the placeholders before sharing it with an agent:

Connect to Audryo at http://localhost:3100/v1.

Use the workspace API key in AUDRYO_API_KEY as a Bearer token. First read
GET /agent_manifest and GET /openapi, then GET /bootstrap. Work only in project
<PROJECT_ID>.

Start by inspecting the project foundation and data context. Propose the most
useful next lifecycle action, explain the evidence, and keep all changes
reviewable. Do not activate, schedule, or send anything without my explicit
approval. Use an Idempotency-Key for every mutation and summarize every change.

Audryo's Use with agent button, available on Audience, Journeys, Campaigns, Emails, Insights, and Data, creates a version of this prompt already scoped to the current project and task.

3. Follow the safe operating loop

1. Discover — read GET /agent_manifest and GET /openapi before making

assumptions about endpoints or delivery policy.

1. Understand — use GET /bootstrap, then read project, foundation, and

data context for the chosen project.

1. Prepare — create reviewable data definitions, audiences, workflows, or

email proposals using stable external IDs and idempotency keys.

1. Validate — dry-run a Journey and inspect every enabled locale before

requesting review.

1. Hand back control — clearly separate what changed, what needs a human

decision, and what would cause external delivery.

A practical first task

For a new product, ask the agent to:

1. Read the product foundation and confirmed data schema.
2. Report missing customer signals without inventing data or consent.
3. Propose one reusable Audience with an explainable rule.
4. Draft one Journey against that Audience.
5. Create an email proposal for each email step and every enabled locale.
6. Run a dry run and return links or IDs for human review.

This produces useful work without sending anything. Once you have reviewed the Journey and confirmed every email variant, use the app or an explicit deliver-scoped API request to activate it.

Guardrails that matter

• Never infer marketing consent. Import proof only when you have it.
• Use Idempotency-Key for every mutation, including test deliveries.
• Do not pass raw customer records into an LLM.
• A contact's preferred language is used only when it is enabled for the

project; otherwise Audryo uses the project's default locale.

• Treat contact erasure as irreversible and require clear human confirmation.
• Read the delivery context before launch. A Journey fails closed until its

sending identity is verified.

4. Prefer MCP, the CLI, or a skill over curl

The HTTP contract is unchanged. For Claude, Cursor, or a terminal, use the packaged client in `packages/audryo` instead of re-deriving every path:

export AUDRYO_API_BASE="http://localhost:3100/v1"
export AUDRYO_API_KEY="sf_live_..."
export AUDRYO_PROJECT_ID="<project-uuid>"

node packages/audryo/bin/audryo.js journeys list
node packages/audryo/bin/audryo.js events ingest events.json
node packages/audryo/bin/audryo.js mcp

Installable skills cover API usage, email authoring, and sending best practices. Details: Agent packaging.

Keep going

• Agent packaging — MCP, CLI, and skills
• Quickstart — authentication, scopes, errors, and the review model
• API introduction — OpenAPI, pagination, discovery
• Contacts & events — sync data safely
• Audiences — preview a dynamic segment before saving it
• Journeys — draft localized emails, dry-run, review, and activate
• Agent API reference — the complete operating contract
• Browse `/llms.txt` for the full doc index