Getting Started with Custom Actions

Stand up your first Custom Action in a few minutes.

Note: Custom Actions require a Builder plan or higher. Upgrade to Builder

Create your first action#

1. Open Custom Actions#

Go to app.avcodex.com, pick your agent, then Build > Capabilities > Add Custom Action.

2. Basic configuration#

Fill in the basics. Example: a tech-facing action that searches your project database for a room.

- Title: Search rooms - Description: Search for a room in the project database by site name, room name, or asset tag - Method: GET - URL: https://api.example.com/rooms

3. Add parameters#

Open the Parameters tab and add a query parameter.

Source: Let the AI generate it.
Key: query
Type: String.
Example: 1200-market boardroom-4
AI instructions: Search term taken from the user's request, usually a site code, room name, or asset tag.
Required: yes.

4. Test the action#

Use the built-in tester to confirm the call works.

5. Save and use#

Click Add Action. Your agent can now search rooms whenever a tech asks.

Import from cURL#

If you already have a cURL command, paste it in. The importer will configure most of the action for you.

1. Paste the cURL#

For example, posting a status note to a Slack channel:

bash

curl -X POST https://api.example.com/messages \
  -H "Authorization: Bearer {{var.API_KEY}}" \
  -H "Content-Type: application/json" \
  -d '{
    "to": "{{channel}}",
    "subject": "{{subject}}",
    "body": "{{message}}"
  }'

2. Parse and configure#

Click Parse cURL to auto-populate:

Method: POST.
Headers: Authorization, Content-Type.
Body parameters with AI placeholders.

3. Configure placeholders#

For each {{placeholder}}:

Add an example value (a real Slack channel, a sample subject like "Boardroom 4 commissioning passed").
Write AI instructions ("Use the channel the tech mentions, default to #av-ops").
Mark required or optional.

Using variables#

Application variables#

Define reusable values in the Variables tab.

Define once:

API_KEY = sk-abc123...
BASE_URL = https://api.example.com

Use anywhere:

URL: {{var.BASE_URL}}/rooms
Header: Authorization: Bearer {{var.API_KEY}}

System variables#

System variables give the action runtime context about the current user and conversation. They are available in every action.

`{{system.message_history}}`

The full conversation as an array.
Format: OpenAI Chat Completions API format.
Use case: send conversation context to an analysis or escalation API.
Example: [{"role": "user", "content": "Q-SYS Core offline"}, {"role": "assistant", "content": "Which room?"}]

`{{system.user_id}}`

The unique ID of the current user.
Format: string identifier.
Use case: track which tech ran an action, attribute work in your PM tool.
Example: "user_abc123"

`{{system.timestamp}}`

Current Unix timestamp, in seconds.
Format: string number.
Use case: log when actions occur, time-stamped commissioning notes.
Example: "1713552052"

Example use in the cURL importer:

bash

curl -X POST https://api.example.com/escalations \
  -H "Content-Type: application/json" \
  -H "X-User-ID: {{system.user_id}}" \
  -d '{
    "timestamp": "{{system.timestamp}}",
    "messages": {{system.message_history}}
  }'

Note: message_history does not need quotes in the body. It is already a JSON array.

Action dependencies#

Action dependencies let you build multi-step workflows where one action's output becomes the next action's input. The agent runs the actions in the right order automatically.

How it works

A tech asks: "What was the last commissioning result for Boardroom 4 at 1200 Market?"
The agent recognizes it needs to:

- First, call Find room to resolve the room ID. - Then, call Get latest commissioning report with that ID.

Actions run in sequence.
The agent responds with the final answer.

Example: two-step room lookup#

Step 1, find the room:

Name: Find room.
URL: https://api.example.com/rooms?query={{query}}.
Returns: {"data": {"id": "room_4421", "name": "Boardroom 4"}}.

Step 2, get the commissioning report:

Name: Get latest commissioning report.
URL: https://api.example.com/rooms/{{roomId}}/commissioning/latest.
Parameter configuration:

- Source: Output from another Action. - Action: Find room. - JSONPath: $.data.id (extracts room_4421 from Step 1).

Result: when the tech asks about Boardroom 4 commissioning, the agent:

Calls Find room, gets room_4421.
Calls Get latest commissioning report with room_4421.
Responds: "Boardroom 4 was commissioned on April 14. All checks passed except the ceiling-mic gate threshold, flagged for review."

JSONPath examples:

$.id - top-level field.
$.data.room.id - nested field.
$.items[0].name - first array item.
$.rooms[*].id - all room IDs.

Verb customization#

Improve the in-chat experience with status messages:

Present tense: "Searching project database."
Past tense: "Searched project database."

These show while the action runs.

Getting Started