AI Module

Integrate Claude (Anthropic) and GPT (OpenAI) into your workflows for text generation, classification, extraction, and summarization — with budget controls and intelligent retry logic.

Module Anthropic Claude OpenAI GPT Token Budget

Supported Providers

Provider	Models	Best For
Anthropic	claude-sonnet-4-20250514 (default), claude-opus-4-20250514, claude-haiku	Complex reasoning, safety, long context
OpenAI	gpt-4o, gpt-4o-mini, gpt-4-turbo	General purpose, vision, function calling

Default Model

If no model is specified in the node configuration, the module uses claude-sonnet-4-20250514 as the default. This provides an excellent balance of quality, speed, and cost.

Configuration

Field	Type	Required	Description
`provider`	string	Yes	`"anthropic"` or `"openai"`
`api_key`	string	Yes	Provider API key (stored encrypted)
`default_model`	string	No	Default model if not specified per-node
`temperature`	float	No	Default temperature (0.0-1.0). Default: `0.7`
`monthly_token_budget`	integer	No	Maximum tokens per month. Default: `1000000`
`max_ai_calls_per_execution`	integer	No	Max AI nodes per single execution. Default: `10`

JSON — Module Configuration Example

{
  "provider": "anthropic",
  "api_key": "sk-ant-api03-xxxxxxxxxxxxxxxxxxxxxxxxx",
  "default_model": "claude-sonnet-4-20250514",
  "temperature": 0.7,
  "monthly_token_budget": 500000,
  "max_ai_calls_per_execution": 5
}

Budget Management

The AI Module includes built-in token budget management to prevent unexpected costs:

Field	Description
`monthly_token_budget`	Maximum total tokens (input + output) allowed per calendar month
`tokens_used_this_month`	Running counter of tokens consumed in the current month
`budget_reset_day`	Day of month when the counter resets (default: 1st)

Budget enforcement works as follows:

Before each AI call, the module checks tokens_used_this_month against monthly_token_budget
If the budget would be exceeded, the node fails with error code budget_exceeded
After each successful call, tokens_used_this_month is incremented by (input_tokens + output_tokens)
On the first day of each month (or configured reset day), a scheduled job resets the counter to 0

Budget Is Per-Organization

The token budget is tracked per organization. If multiple workflows in the same organization use AI nodes, they share the same budget pool. Monitor usage in Settings → Modules → AI → Usage.

Rate Limiting

The AI Module applies two layers of rate limiting:

Limit	Scope	Default	Purpose
Calls per hour	Per organization	60	Prevent runaway workflows from draining budget
Calls per execution	Per workflow execution	10	Prevent infinite loops with AI nodes

When either limit is hit, the node fails with error code rate_limited and the execution step records which limit was exceeded.

Node Type: action.ai.prompt

Config Field	Type	Required	Description
`prompt`	string	Yes	User prompt text (supports variable interpolation)
`system_prompt`	string	No	System/context prompt to set AI behavior
`model`	string	No	Override default model for this node
`temperature`	float	No	Override default temperature (0.0-1.0)
`max_tokens`	integer	No	Maximum output tokens. Default: `1024`

JSON — AI Node Configuration Example

{
  "type": "action.ai.prompt",
  "config": {
    "system_prompt": "You are a customer support classifier. Categorize the message into: billing, technical, general, or urgent. Respond with only the category name.",
    "prompt": "Classify this customer message:\n\n{{trigger.message}}",
    "model": "claude-sonnet-4-20250514",
    "temperature": 0.1,
    "max_tokens": 50
  }
}

The node output contains:

JSON — Node Output

{
  "response": "technical",
  "model": "claude-sonnet-4-20250514",
  "input_tokens": 45,
  "output_tokens": 3,
  "total_tokens": 48
}

Retry Logic

The AI Module implements exponential backoff for transient errors:

Attempt	Delay	Applies To
1st retry	1 second	rate_limited, timeout, server_error (5xx)
2nd retry	2 seconds	rate_limited, timeout, server_error (5xx)
3rd retry	4 seconds	rate_limited, timeout, server_error (5xx)

Never Retried

The error content_rejected (content policy violation) is never retried. The same prompt will produce the same rejection. The node fails immediately with a clear error message.

Error Codes

Error Code	Description	Retryable	Resolution
`rate_limited`	Provider API rate limit hit	Yes	Wait and retry (automatic with backoff)
`context_exceeded`	Input exceeds model's context window	No	Reduce prompt length or use a model with larger context
`content_rejected`	Content policy violation	No	Modify prompt to comply with provider's usage policy
`timeout`	Request took longer than 60 seconds	Yes	Reduce max_tokens or simplify prompt
`budget_exceeded`	Monthly token budget exhausted	No	Increase budget or wait for monthly reset
`invalid_api_key`	API key rejected by provider	No	Check and update API key in module configuration

Common Issues

Problem	Cause	Solution
API key invalid	Key expired, revoked, or incorrectly entered	Generate a new key from the provider dashboard. Ensure no leading/trailing whitespace.
Budget exceeded	Monthly token budget consumed	Increase `monthly_token_budget` in module config, or wait for the monthly reset.
Context too long	Prompt + variable interpolation exceeds model context	Truncate large inputs before passing to AI. Use a model with larger context (e.g., Claude with 200k).
Slow responses	High max_tokens or complex prompt	Reduce `max_tokens`. Use a faster model (e.g., claude-haiku or gpt-4o-mini).
Content rejected	Prompt or output violates usage policy	Rephrase prompt. Add guardrails in system_prompt. This error is never retried.