Lens Node

Type: lens · Category: Lens

Execute a LenserFight lens prompt with model, funding, and parameter overrides. The Lens node is the core AI execution unit — it takes text input, runs it through a selected lens prompt with a configured model, and returns generated text.

---

When to Use

Use the Lens node when you want to run a specific lens (a versioned prompt template) as a workflow step. Lens nodes are ideal for:

• Generating summaries, reports, or analyses from upstream data
• Running AI evaluation or judgement with a crafted rubric prompt
• Applying domain-specific transformations using your own published lenses
• Chaining multiple lens executions in sequence or parallel

---

Inputs

Name	Type	Required	Description
text	text	Yes	Text from an upstream node (e.g. a JSON transform, RAG retriever, or prompt template).

Also accepts json and object inputs.

---

Outputs

Name	Type	Shape
text	text	{ text: text, modelId: text }

---

Required Config

Field	Type	Description
model_id	string	AI model key to use (e.g. openai:gpt-4.1-mini, anthropic:claude-3-7-sonnet).

---

Optional Config

Field	Type	Default	Description
param_overrides	json	—	Lens parameter overrides — key/value pairs injected into the lens template variables.
funding_source	select	platform_credit	platform_credit · user_byok_cloud · user_byok_local

---

Example

{
  "model_id": "openai:gpt-4.1-mini",
  "funding_source": "platform_credit",
  "param_overrides": { "tone": "concise" }
}

Scenario: Generate the weekly arena digest from aggregated battle data.

Expected input: "Summarize this week of arena results."

Expected output: "Weekly digest with top contenders, outliers, and recommended rematches."

Downstream: → email_send with { "body": "$.text" }

---

Valid Connections

→ email_send, slack_notify, discord_notify — for delivery of generated text

→ output_parser — to extract structured fields from generated JSON

→ judge_evaluator — to evaluate lens outputs against a rubric

→ memory_write — to persist generated content

→ logger — to audit lens output

---

Invalid Connections

✗ Cannot be the source for another Lens node's text input unless the upstream text is reformatted — Lens output is plain text, not structured JSON.

✗ Cannot connect directly to embedding without routing through text_splitter first for long outputs.

---

Execution Notes

• Lens nodes always run through the versioned lens execution engine — parameter overrides are layered on top of the lens's published defaults.
• The model_id must be available for the chosen funding_source. Local gateway models require user_byok_local.
• Token limits are enforced per model. For long inputs, use text_splitter → embedding → rag_retrieval before the Lens node.

---

Troubleshooting

• "Model not available" — verify that model_id is in scope for your workspace's enabled providers.
• "Parameter override not applied" — check that param_overrides keys match the lens template variable names exactly (case-sensitive).
• "Empty output" — the upstream text input may be empty; add a logger or debug_inspector node before the Lens node to inspect the payload.

---

See also: AI Primitives — Lens Execute · Node Catalog Index · Workflow Studio · Execution Engine