A Proxy to extract structured data from text and images using LLMs.
l1m is the easiest way to get structured data from unstructured text or images using LLMs. No prompt engineering, no chat history, just a simple API to extract structured json from text or images.
curl -X POST https://api.l1m.io/structured \
-H "Content-Type: application/json" \
-H "X-Provider-Url: demo" \
-H "X-Provider-Key: demo" \
-H "X-Provider-Model: demo" \
-d '{
"input": "'$(curl -s https://public.l1m.io/menu.jpg | base64)'",
"schema": {
"type": "object",
"properties": {
"items": {
"type": "array",
"items": {
"type": "object",
"properties": {
"name": { "type": "string" },
"price": { "type": "number" }
}
}
}
}
}
}'↑ Copy and run this example in your terminal. The demo endpoints return pre-rendered LLM responses for quick testing.
curl -X POST https://api.l1m.io/structured \
-H "Content-Type: application/json" \
-H "X-Provider-Url: demo" \
-H "X-Provider-Key: demo" \
-H "X-Provider-Model: demo" \
-d '{
"input": "A particularly severe crisis in 1907 led Congress to enact the Federal Reserve Act in 1913",
"schema": {
"type": "object",
"properties": {
"items": {
"type": "array",
"items": {
"type": "object",
"properties": {
"name": { "type": "string" },
"price": { "type": "number" }
}
}
}
}
}
}'↑ Copy and run this example in your terminal. The demo endpoints return pre-rendered LLM responses for quick testing.
#!/bin/bash
# Run the same request multiple times with caching enabled
# A cache key is generated from the input, schema, provider key and model
# Cache key = hash(input + schema + x-provider-key + x-provider-model)
for i in {1..5}; do
curl -X POST https://api.l1m.io/structured \
-H "Content-Type: application/json" \
-H "X-Provider-Url: $PROVIDER_URL" \
-H "X-Provider-Key: $PROVIDER_KEY" \
-H "X-Provider-Model: $PROVIDER_MODEL" \
-H "X-Cache-TTL: 300" \
-d '{
"input": "The weather in San Francisco is sunny and 72°F",
"schema": {
"type": "object",
"properties": {
"temperature": { "type": "number" },
"conditions": { "type": "string" }
}
}
}'
echo "\n--- Request $i completed ---\n"
done
# Only 1 LLM call will be made, subsequent calls will be served from cache
# for the duration specified in X-Cache-TTL (300 seconds in this example)
While l1m does not support "tool calling" in the same sense as other model providers, you can achieve similar
results using the enum property.
INPUT="Please find the user [email protected] and get their details"
# First call to determine which tool to use
TOOL_RESPONSE=$(curl -X POST https://api.l1m.io/structured \
-H "Content-Type: application/json" \
-H "X-Provider-Url: $PROVIDER_URL" \
-H "X-Provider-Key: $PROVIDER_KEY" \
-H "X-Provider-Model: $PROVIDER_MODEL" \
-d '{
"input": "$INPUT",
"instruction": "Select the most appropriate tool based on the input",
"schema": {
"type": "object",
"properties": {
"selected_tool": {
"type": "string",
"enum": ["getUserByEmail", "getUserByName", "getUserById"]
}
}
}
}')
# Extract the selected tool from the response
SELECTED_TOOL=$(echo $TOOL_RESPONSE | jq -r '.selected_tool')
# Switch case to handle different tool types and extract appropriate arguments
case $SELECTED_TOOL in
"getUserByEmail")
# Make a follow up call to extract additional argument
ARGS=$(curl -X POST https://api.l1m.io/structured \
-H "Content-Type: application/json" \
-H "X-Provider-Url: $PROVIDER_URL" \
-H "X-Provider-Key: $PROVIDER_KEY" \
-H "X-Provider-Model: $PROVIDER_MODEL" \
-d '{
"input": "$INPUT",
"schema": {
"type": "object",
"properties": {
"email": {
"type": "string",
"format": "email",
"description": "Extract the email address to use as argument"
}
}
}
}')
EMAIL=$(echo $ARGS | jq -r '.email')
echo "Calling getUserByEmail with email: $EMAIL"
;;
"getUserByName")
# ...
"getUserById")
# ...
*)
echo "Error: Unknown tool selected: $SELECTED_TOOL"
exit 1
;;
esac#!/bin/bash
# Make sure Ollama is running locally with your desired model
# Example: ollama run llama3
# Set up ngrok to expose your local Ollama server
# ngrok http 11434 --host-header="localhost:11434"
# Replace the ngrok URL with your actual ngrok URL or use localhost
# if you're running this on the same machine as Ollama
curl -X POST https://api.l1m.io/structured \
-H "Content-Type: application/json" \
-H "X-Provider-Url: https://your-ngrok-url.ngrok-free.app/v1" \
-H "X-Provider-Key: ollama" \
-H "X-Provider-Model: llama3:latest" \
-d '{
"input": "A particularly severe crisis in 1907 led Congress to enact the Federal Reserve Act in 1913",
"schema": {
"type": "object",
"properties": {
"year": {
"type": "number",
"description": "The year of the federal reserve act"
}
}
}
}'x-provider-model (optional): LLM model to usex-provider-url (optional): LLM provider URL (See supported providers below)
x-provider-key (optional): API key for LLM provider
x-cache-ttl (optional): Cache TTL in seconds.input String or base64 image data.
schema A JSON Schema-like object describing the desired output object.
object.
array and enum are supported forf nested properties.
instruction Optional instruction to include in the request to the model.
data.
For images, base64 encoded data can be in one of the following formats:
image/jpegimage/pngl1m provides official SDKs for multiple programming languages to help you integrate structured data extraction into your applications: