¶ Description
Text completion with an OpenAI model. This node reads a prompt column from the input table and writes model outputs to new columns (prefixed).
¶ Parameters
¶ Parameters tab
Parameters:
- Column with test to expand on
- API Key
- Parameter model
- Parameter temperature
- Parameter max_tokens
- Parameter top_p
- Parameter frequency penalty
- Parameter presence_penalty
- Parameter n
- Prefix output column name
- Debug information level
- Optional: extra parameters for cURL
- Number of retries on connection error
¶ Description tab
See dedicated page for more information.
¶ Code tab
DaVinci is a scripted action. Embedded code is accessible and customizable through this tab.
¶ Configuration tab
See dedicated page for more information.
¶ About
Text completion with an OpenAI model. This node reads a prompt column from the input table and writes model outputs to new columns (prefixed).
¶ What this node does
- Takes one text column as prompt (e.g.,
C1) - Calls OpenAI’s Completions/Chat API with your parameters (model, temperature, max tokens, etc.)
- Appends new columns to the table (e.g.,
openai_text, plus optional meta), then passes the table to the next node
¶ Prerequisites
- An OpenAI API key with active prepaid credits (or a project spending limit that isn’t exhausted).
- An upstream table with at least one column holding prompts (e.g., via
InlineTableorreadCSV).
¶ How to get an OpenAI API key (step-by-step)
-
Sign in to the OpenAI Platform and go to API keys.
-
Click Create new secret key.
-
(If prompted) Choose the Project; give the key a name; Permissions: All.
-
Click Create secret key → Copy the key (starts with
sk-…orsk-proj-…). You won’t see it again.
-
(If you get insufficient_quota later) Open Billing and buy credits / raise the project’s spending limit.
- Paste the key into daVinci → API Key.
Security: Treat the key as a secret. Don’t commit it to version control or share it in screenshots.
¶ Minimal input (copy-paste CSV)
Create a CSV and feed it via readCSV (or build the rows with InlineTable):
C1
"Write a two-sentence tagline for a coffee shop called 'Bean Scene'."
"Summarize in one sentence: The board approved a budget increase to expand the mobile app team next quarter."
"List 3 bullet points (short) describing benefits of daily walking."
Connect that node’s output to daVinci.
¶ Parameter reference (what to enter)
| UI Field (Id) | Meaning / Notes | Example / Default |
|---|---|---|
| Column with text to expand on | Column containing the prompt text. | C1 |
| API Key | Your OpenAI API key (sk-…). |
sk-… |
| Parameter model | Model name string. | gpt-3.5-turbo (safe default) |
| Parameter temperature | Temperature (creativity). 0–2. | 1 |
| Parameter max_tokens | max_tokens—upper bound for output length. |
96 (raise for longer) |
| Parameter top_p | top_p nucleus sampling. 0–1. |
1 |
| Parameter frequency_penalty | frequency_penalty |
0 |
| Parameter presence_penalty | presence_penalty |
0 |
| Parameter n | Number of completions per row. | 1 |
| Prefix of output column name | Prefix for output columns. | openai_ |
| Debug information level | Verbosity in logs. | nothing |
| Optional: extra parameters for cURL | Extra cURL args (normally leave empty). | (empty) |
| Number of retries on connection error | Retries on connection error. | 10 |
Model tip: If available in your account,
gpt-4o-miniis a cost-efficient upgrade. Use any model string supported by your OpenAI project.
¶ Quick start (works-first-try)
-
Upstream node outputs a column
C1with prompts. -
In daVinci → Parameters:
- Column with text to expand on =
C1 - API Key = your
sk-…key - Parameter model =
gpt-3.5-turbo - Parameter max_tokens =
96(or256if you need more text) - Leave other settings as shown above
- Column with text to expand on =
-
Run the pipeline.
-
Open Data after daVinci: you should see columns like
openai_textcontaining the model’s responses.
¶ Output columns
<prefix>text— the generated text (e.g.,openai_text)- Optional metadata columns may be added depending on node version (tokens, finish reason, etc.), all starting with your prefix.
¶ Tuning & best practices
- Deterministic output: set temperature to
0–0.2. - Longer output: raise max_tokens (watch costs).
- Multiple options: set Parameter n > 1 (you’ll get additional columns/variants).
- Prompting: include instructions in the input column (style, length, JSON-only, etc.).
- PII: avoid sending sensitive data; mask where possible.
¶ Common errors & fixes
| Error in Log | What it means | Fix |
|---|---|---|
insufficient_quota |
No credits / hit spending cap | Buy credits / raise project limit; retry |
invalid_api_key / 401 |
Key wrong or revoked | Create a new key in the correct project and paste into API Key |
model_not_found |
Model not available to your project | Use a supported model (e.g., gpt-3.5-turbo, gpt-4o-mini) |
rate_limit_exceeded |
Too many requests | Lower concurrency / add backoff; keep Number of retries on connection error |
context_length_exceeded / max_tokens |
Prompt + max_tokens too long | Shorten prompt or reduce max_tokens |
¶ Example pipeline (recommended)
readCSV → daVinci → writeCSV
readCSVreads the sample abovedaVincigeneratesopenai_textwriteCSVsaves results for download
