Use Future AGI in n8n
Retrieve and compile Future AGI prompts in n8n, or configure Preview OpenTelemetry execution tracing.
Use the Future AGI community node to retrieve and compile prompts in an n8n workflow.
To send traces of your n8n executions to Future AGI instead, see Trace n8n executions with OpenTelemetry.
Install the prompt-management node
Install the Future AGI community node to retrieve prompts. This node is separate from native execution tracing below.
- Access Community Nodes: Navigate to the Community Nodes section in your n8n instance settings.
Open Community nodes in n8n Settings.
- Install Future AGI Node: Enter the Future AGI community node package name:
@future-agi/n8n-nodes-futureagi
The Community nodes page provides the installation action.
- Verify Installation: Confirm the Future AGI node is successfully installed and available in your workflow editor.
Enter @future-agi/n8n-nodes-futureagi in the install dialog.
Configure the node
Add the node to your workflow, then connect your Future AGI account.
Add the node to your workflow
Once you have installed the community node, you can start using it in your workflows:
- Search for Future AGI Node: In your n8n workflow editor, search for “Future AGI” in the list of available nodes.
The workflow editor opens the node picker beside the canvas.
- Select the Node: Click on the Future AGI node to add it to your workflow.
Search for Future AGI in the node picker.
- View Available Operations: Once added, click on the node to view the list of supported operations:
The Future AGI node offers a Get prompt action.
Supported Operations:
- Get a prompt
- Select Operation: Click on your desired operation to proceed to the configuration screen.
The Get Prompt operation includes version and compilation controls.
Set up credentials
Once you’re on the operation screen, you’ll need to configure your Future AGI credentials:
- Access Settings Window: Navigate to the settings window to add your API credentials.
The credential form contains the base URL and masked key fields.
- Configure API Credentials: Enter your Future AGI credentials:
- Base URL:
https://api.futureagi.com - API Key: Your Future AGI API key
- Secret Key: Your Future AGI secret key
- Base URL:
Get your API key and secret key from API keys.
Future AGI’s API keys page masks key values.
- Test the connection before saving the credentials.
Retrieve a prompt
Select a prompt in the node to retrieve it in your workflow.
Choose an available prompt from the node dropdown.
The Future AGI Prompts page lists saved prompts.
Choose a prompt version
You can use this node to fetch whatever version of the prompt you have selected. By default, the node will retrieve the default version of your chosen prompt.
Use Default Version is enabled and the node output contains the selected prompt.
If you want to select a specific version of the prompt instead of using the default, you can do so by switching off the “Use Default Prompt” toggle. This gives you full control over which version of your prompt to use in your workflow, allowing you to test different iterations or use specific versions for different use cases.
Turning off Use Default Version exposes the Prompt Version field.
Compile prompt variables
Future AGI follows the convention of using words inside double curly brackets (i.e., words in {{}}) as variables in prompts. To replace these variables with actual values, users can use the Use compiled prompt option of the node.
When you toggle this switch on, a code editor will appear where you need to write a JSON object that maps the variables to their values like key-value pairs.
Important Note: The keys must be exactly the same as the variables in your prompt, as it is case sensitive.
Use Compiled Prompt exposes a JSON variable map.
Trace n8n executions with OpenTelemetry
Use n8n’s built-in OpenTelemetry exporter to send workflow and node execution traces to Future AGI. This path does not use the prompt-management community node.
Warning
n8n labels OpenTelemetry tracing Preview and advises against relying on it in production workflows because it may change between releases.
| n8n version | 2.19.0 or later for workflow and node spans; 2.33.0 or later for agent spans. n8n labels its OpenTelemetry tracing Preview. |
| n8n edition | Self-hosted. On n8n Cloud, OpenTelemetry availability depends on n8n’s plan and version support; check whether your instance exposes Settings > OpenTelemetry. |
| Transport | OTLP/HTTP with protobuf (n8n’s default) to https://api.futureagi.com/tracer/v1/traces, uncompressed |
| Project | The X-Project-Name header, or the project_name resource attribute |
| Future AGI collector | v1.58.0 or later for project-header routing. With an older collector, use the resource-attribute alternative below. |
| Reference | n8n OpenTelemetry tracing guide |
1. Configure tracing
Choose the Settings route or the environment-based alternative below.
Settings > OpenTelemetry
From n8n 2.27.0, an instance owner or admin can configure tracing in Settings > OpenTelemetry. Turn on Enable OpenTelemetry and enter:
| Field | Value |
|---|---|
| OTLP endpoint | https://api.futureagi.com |
| Trace path | /tracer/v1/traces |
| Protocol | http/protobuf |
| Headers | X-Api-Key=<your-api-key>,X-Secret-Key=<your-secret-key>,X-Project-Name=n8n |
Get your keys from API keys. For a self-hosted Future AGI, replace the endpoint with its base URL.
Select Save settings. n8n applies the settings without a process restart. The X-Project-Name=n8n header routes traces to the n8n project in your API key’s workspace, creating an Observe project if it doesn’t exist. You do not need OTEL_RESOURCE_ATTRIBUTES for this route.
Environment variables take precedence: a field whose variable is set is disabled in the form. Leave the corresponding N8N_OTEL_* variables unset to manage those fields in Settings.
Environment variables (self-hosted alternative)
Set these on every n8n process: the main instance and, in queue mode, every worker and webhook processor. Then restart n8n.
export N8N_OTEL_ENABLED=true
export N8N_OTEL_EXPORTER_OTLP_ENDPOINT=https://api.futureagi.com
export N8N_OTEL_EXPORTER_OTLP_TRACING_PATH=/tracer/v1/traces
export N8N_OTEL_EXPORTER_OTLP_HEADERS="X-Api-Key=<your-api-key>,X-Secret-Key=<your-secret-key>"
export OTEL_RESOURCE_ATTRIBUTES="project_name=n8n,project_type=observe"
N8N_OTEL_EXPORTER_OTLP_ENDPOINTis the base URL. n8n appendsN8N_OTEL_EXPORTER_OTLP_TRACING_PATH, which defaults to/v1/traces, so set it to Future AGI’s path. For a self-hosted Future AGI, use its base URL instead.- Get the keys from API keys. To keep them out of the environment, put the header string in a file and point
N8N_OTEL_EXPORTER_OTLP_HEADERS_FILEat it. - This alternative uses
OTEL_RESOURCE_ATTRIBUTESto setproject_name=n8nandproject_type=observe. A non-emptyproject_nameresource attribute takes precedence overX-Project-Name. Future AGI creates the project in your API key’s workspace on the first export if it doesn’t exist. A resource with neither aproject_nameattribute nor anX-Project-Nameheader is rejected with HTTP 400. - Leave
N8N_OTEL_EXPORTER_OTLP_PROTOCOLat its default,http/protobuf.
2. Run a production execution
By default n8n only exports production executions: runs of a published workflow started by its trigger, such as a webhook call, a schedule or a chat message. Runs you start from the editor are not exported. To export them while you test, disable production-only filtering in Settings, or set:
export N8N_OTEL_TRACES_PRODUCTION_ONLY=false
Trigger the workflow, wait a few seconds for the batch to flush, then open your project in Observe.
Warning
Send test trace does not include OTEL_RESOURCE_ATTRIBUTES. With resource-only routing, the test therefore has neither a project_name resource attribute nor an X-Project-Name header and is rejected with HTTP 400, even when real executions export correctly. Add X-Project-Name=n8n to the exporter headers to use Send test trace with a collector that supports project headers, or validate the resource-only setup by triggering a workflow. The Settings configuration above already includes this header.
What you get
The model-child coverage below applies to n8n 2.45.7 with its bundled @n8n/agents 0.30.6. It uses n8n’s native AI SDK telemetry and the same exporter as workflow tracing; you do not need another tracing package. Earlier releases with agent tracing may have different span coverage.
| Span | One per | Attributes |
|---|---|---|
workflow.execute | Execution | Workflow ID, name and version, execution ID, mode and status, error type, n8n project ID |
node.execute | Node run, nested under its workflow span | Node name, type and version, input and output item counts |
<agent name>.generate or <agent name>.stream | Agent run (the Agents feature) | gen_ai.operation.name=invoke_agent, agent name, model, thread ID, tool catalog in gen_ai.prompt |
ai.generateText or ai.streamText | AI SDK call within the agent run | Model metadata and aggregate ai.usage.inputTokens / ai.usage.outputTokens |
ai.generateText.doGenerate or ai.streamText.doStream | Model step, nested under its AI SDK call | gen_ai.request.model, gen_ai.response.model, gen_ai.usage.input_tokens, gen_ai.usage.output_tokens; model inputs and responses when content capture is enabled |
execute_tool <tool name> | Tool call an agent makes | Tool name and call ID; arguments and result when content capture is enabled |
Execution spans share a resource with service.name=n8n, the n8n version and instance ID, and the host and process details that n8n’s resource detectors add. Those include the host name, the OS user running n8n and its command line. The environment-based setup also supplies project_name and project_type; for the header-based setup, Future AGI fills the missing project_name from X-Project-Name during ingestion.
- Workflow and node spans carry metadata only: names, IDs, item counts and status. The items themselves are not exported. They appear as generic spans in Observe, not LLM or tool spans; the native exporter does not assign Future AGI span kinds.
- The older AI Agent (LangChain) node produces only a
node.executespan, like any other node. Its prompts, model calls and tool calls are not traced. Agent spans come only from n8n’s newer Agents feature. - For token counts, inspect the model child spans in Observe, not just the agent root. Usage is available when the model provider returns it. Automatic LLM/tool classification, token summaries and cost display depend on Future AGI’s attribute mapping and model pricing; raw token attributes alone are not a cost estimate.
- AI SDK parent spans repeat usage for their model children. When aggregating usage, count the model child spans once; do not add parent totals or count
ai.usage.*andgen_ai.usage.*as separate usage. Avoid adding a second model instrumentor alongside n8n’s native tracing.
Turn off model and tool content
Agent tracing records model inputs and responses, tool arguments and tool results by default. Set both flags to disable those payloads while retaining model and token metadata:
export N8N_AGENTS_TRACING_RECORD_INPUTS=false # model inputs and tool arguments
export N8N_AGENTS_TRACING_RECORD_OUTPUTS=false # model responses and tool results
These flags are not a general redaction filter: names, IDs, tool definitions and error details can still contain sensitive data, as can the host and process resource attributes described above. If your policy requires excluding agent telemetry entirely, set N8N_AGENTS_TRACING_ENABLED=false; workflow and node spans remain.
Questions & Discussion