llm-1minai

Plugin for LLM adding support for 1min.ai AI models.

Access 75+ AI models through a single API: Claude 4.6 Sonnet/Opus, Claude 4.5 Haiku, GPT-5.4, GPT-5.1 Codex, o3 / o4, Gemini 3.1, Qwen3, DeepSeek V3.2, Mistral, LLaMA 4, Grok 4, Sonar, and more.

Built on the unified /api/chat-with-ai endpoint with SSE streaming, image and file attachments, AI memory across conversations, brand voice, and structured prompt settings.

Looking for a standalone CLI? omi is a single-binary Go port of this plugin. No Python required, sub-20ms startup, same API-KEY and same routing (UNIFY_CHAT_WITH_AI for chat / vision / docs, CODE_GENERATOR + SPEECH_TO_TEXT on /api/features). Run go install github.com/gl0bal01/omi/cmd/omi@latest.

Installation

Using LLM's Plugin Manager (Recommended)

Install this plugin in the same environment as LLM:

llm install llm-1minai

Using pipx

If you installed LLM with pipx, inject the plugin into the LLM environment:

pipx inject llm llm-1minai

Or install both LLM and the plugin with pipx:

pipx install llm
pipx inject llm llm-1minai

From Source (Development)

Install from source for development:

git clone https://github.com/gl0bal01/llm-1minai
cd llm-1minai
llm install -e .

Or with pipx:

git clone https://github.com/gl0bal01/llm-1minai
cd llm-1minai
pipx inject llm -e .

Upgrade or Uninstall

Upgrade an existing install

# Plugin-manager install
llm install -U llm-1minai

# pipx-injected install
pipx runpip llm install -U llm-1minai

# Editable / source install
cd llm-1minai && git pull && llm install -e .

Uninstall

# Plugin-manager install
llm uninstall llm-1minai

# pipx-injected install
pipx uninject llm llm-1minai

# Editable / source install
pip uninstall llm-1minai

Verify removal

llm models list | grep 1min   # should print nothing
llm plugins | grep 1min       # should print nothing

Wipe local data (optional)

The plugin keeps user settings and the conversation-UUID map at ~/.config/llm-1min/. Remove them after uninstall if you want a clean slate:

rm -rf ~/.config/llm-1min/

The 1min.ai API key is stored separately by LLM, not under that directory:

llm keys remove 1min
unset ONEMIN_API_KEY  # if you exported it

Migrating to the standalone Go CLI

If you're switching to omi, it uses a different config dir (~/.config/omi/) and won't conflict with leftover Python-plugin state. After uninstalling above:

go install github.com/gl0bal01/omi/cmd/omi@latest
omi config set api_key sk-...

Configuration

You need a 1min.ai API key to use this plugin. Get one from 1min.ai.

Set your API key using:

llm keys set 1min
# Paste your API key when prompted

Or set it as an environment variable:

export ONEMIN_API_KEY="your-api-key-here"

Available Models

Quick View

See all 1min.ai models in your terminal:

# Show all LLM models (look for "1min.ai:" prefix)
llm models list | grep "1min.ai"

# Or use our dedicated command with descriptions
llm 1min models

Model Categories

75+ models across 10 providers (v0.4.0):

• OpenAI (~25 models): GPT-3.5/4/4.1/4o, GPT-5, GPT-5.1, GPT-5.1 Codex / Codex Mini, GPT-5.2 / 5.2 Pro, GPT-5.4 / Mini / Nano / Pro, o3 / o3 Pro / o3 Deep Research, o4 Mini / o4 Mini Deep Research
• Anthropic (8 models): Claude 4 / 4.5 / 4.6 Sonnet, Claude 4 / 4.1 / 4.5 / 4.6 Opus, Claude 4.5 Haiku
• Google (5 models): Gemini 2.5 Flash / Pro, Gemini 3 Flash / 3.1 Flash Lite / 3.1 Pro (Preview)
• Alibaba (Qwen) (10 models): Qwen3 Max / VL Plus / VL Flash / Coder Plus / Coder Flash, Qwen Max / Plus / Flash / VL Max / VL Plus
• DeepSeek (2 models): DeepSeek V3.2 Chat, DeepSeek V3.2 Reasoner
• xAI (6 models): Grok 3 / 3 Mini, Grok 4, Grok 4 Fast Reasoning / Non-Reasoning, Grok Code Fast 1
• Mistral (7 models): Mistral Small / Medium 3.1 / Large 2, Magistral Small / Medium 1.2, Ministral 14B, Open Mistral Nemo
• Meta / open-source (6 models): LLaMA 2 70b, LLaMA 3 70b, LLaMA 4 Scout / Maverick, GPT OSS 20b / 120b
• Cohere (1 model): Command R
• Perplexity (4 models): Sonar, Sonar Pro, Sonar Reasoning Pro, Sonar Deep Research

Important: Use llm 1min models to see all available models with descriptions.

For a complete model reference with IDs and usage examples, see MODEL_SELECTION.md.

Usage

List Available Models

First, see what models are available:

# See model IDs and friendly names
llm 1min models

# Or see just the model IDs
llm models list | grep "1min.ai"
# Output example:
# 1min.ai: gpt-4o-mini          ← Use this ID with -m flag
# 1min.ai: claude-sonnet-4-20250514

Basic Usage

Use the model ID (shown after "1min.ai:") with the -m flag:

# Fast and efficient - GPT-4o Mini
llm -m 1min/gpt-4o-mini "Explain quantum computing in simple terms"

# Most powerful - Claude 4 Opus or GPT-5
llm -m 1min/claude-4-opus "Design a complex system architecture"
llm -m 1min/gpt-5 "Advanced reasoning task"

# Best for coding - Claude 4 Sonnet
llm -m 1min/claude-4-sonnet "Write a REST API with FastAPI"

# Best for reasoning - O4 Mini or DeepSeek R1
llm -m 1min/o4-mini "Solve this logic puzzle"
llm -m 1min/deepseek-reasoner "Complex math problem"

# Web-aware - Sonar or Sonar Reasoning
llm -m 1min/sonar "What are the latest AI developments?"
llm -m 1min/sonar-reasoning-pro "Research topic with citations"

# Fast responses - Claude 4.5 Haiku or Gemini Flash
llm -m 1min/claude-4-5-haiku "Quick question"
llm -m 1min/gemini-2.5-flash "Fast response needed"

Debug Mode (Troubleshooting)

See exactly what's being sent to the API:

# Enable debug to see all options and API payload
llm -m 1min/gpt-4o -o debug true "test prompt"

# Or use environment variable
LLM_1MIN_DEBUG=1 llm -m 1min/gpt-4o "test prompt"

# See all available options for any model
llm models --options | grep -A 8 "1min/gpt-4o"

Note: The -d flag is already used by LLM for database operations, so debug uses -o debug true. Debug output redacts prompt text, attachment keys/IDs, and brand voice IDs by default. For more details, see DEBUG_USAGE.md

Conversation Mode (Chat)

Continue a conversation with context using the -c flag:

# Start a conversation
llm -m 1min/gpt-4o "Hello, I need help with Python"

# Continue the conversation (uses -c flag)
llm -m 1min/gpt-4o -c "What are list comprehensions?"

# Keep going - model is remembered automatically
llm -c "Show me an example"

# Resume a specific conversation by ID
llm -c --cid <conversation-id> "Continue this topic"

# View conversation history
llm logs -n 5

Important: The -c flag is handled by the LLM framework, not this plugin. Each continuation re-sends prior messages to maintain context.

See the LLM chat documentation for more details.

Code Generation

Code-focused models automatically use CODE_GENERATOR mode by default:

# These models auto-use CODE_GENERATOR (no -o needed)
llm -m 1min/claude-4-6-sonnet "Create a REST API with FastAPI"
llm -m 1min/qwen3-coder-plus "Create a simple REST API with Go"
llm -m 1min/grok-code-fast-1 "Optimize this algorithm"
llm -m 1min/deepseek-reasoner "Refactor this function"

# Explicitly set for other models
llm -m 1min/gpt-4o \
  -o conversation_type CODE_GENERATOR \
  "Write a binary search function"

Built-in defaults (see llm 1min options defaults):

• Code models: claude-sonnet-4-6, claude-opus-4-6, claude-haiku-4-5-20251001, qwen3-coder-plus, qwen3-coder-flash, grok-code-fast-1, gpt-5.1-codex, gpt-5.1-codex-mini, deepseek-reasoner → CODE_GENERATOR
• Web-aware models: sonar, sonar-pro, sonar-reasoning-pro, sonar-deep-research, o3-deep-research, o4-mini-deep-research → web_search=true

Advanced Options

Available Options

• conversation_type: UNIFY_CHAT_WITH_AI (default) or CODE_GENERATOR
• web_search: Enable real-time web search (true/false, default: false)
• num_of_site: Number of sites to search when web_search is enabled (1-10, default: 3)
• max_word: Maximum words from web search results (100-10000, default: 1000)
• history_mixed: Mix context between different models (true/false, default: false) (renamed from is_mixed in v0.4.0; run llm 1min options migrate to update saved configs)
• history_limit: Max history messages included as context (1-50, default: 10)
• with_memories: Enable AI memory across conversations (true/false, default: false)
• brand_voice_id: Brand voice ID for response style (string, default: none)
• images: Comma-separated image asset keys from Asset API (string, default: none)
• files: Comma-separated file IDs from Asset API (string, default: none)
• debug: Show detailed API request information (true/false, default: false)
  • Use: -o debug true (Note: -d is taken by LLM's database option)
  • See: DEBUG_USAGE.md for details

One-Time Usage (CLI Flags)

# Enable web search for any model
llm -m 1min/gpt-4o \
  -o web_search true \
  -o num_of_site 5 \
  "What's the latest in AI?"

# Use code generator mode
llm -m 1min/gpt-4o \
  -o conversation_type CODE_GENERATOR \
  "Write a binary search algorithm"

# Mix context between models
# Start a conversation first (no -c needed for the first turn)
llm -m 1min/gpt-4o "My name is Fabien"

# Continue with a different model: -c reuses the LLM conversation,
# history_mixed=true shares the 1min.ai conversation UUID across models.
llm -m 1min/claude-4-6-opus -c -o history_mixed true "What is my name?"

# Stream responses (SSE) - streaming is the default for `llm prompt`
llm -m 1min/gpt-4o "Explain transformers in 200 words"

# Disable streaming with --no-stream
llm -m 1min/gpt-4o --no-stream "Explain transformers in 200 words"

# Image attachments — upload first to get an asset key
KEY=$(llm 1min upload -q photo.png)
llm -m 1min/gpt-4o -o images "$KEY" "What do you see in this image?"

# Or upload interactively (prints key + usage hint)
llm 1min upload photo.png

# Cross-conversation memory
llm -m 1min/gpt-4o -o with_memories true "Remember I prefer Python over JS"

# Debug mode - see what's being sent to the API
llm -m 1min/gpt-4o -o debug true "test prompt"

# Alternative: environment variable
LLM_1MIN_DEBUG=1 llm -m 1min/gpt-4o "test prompt"

# Or set as default (useful for troubleshooting)
llm 1min options set debug true

Persistent Configuration

Set default options that apply automatically:

# Set global defaults
llm 1min options set web_search true
llm 1min options set num_of_site 5

# Now web search is enabled by default
llm -m 1min/gpt-4o "Latest AI news"  # Uses web_search=true automatically

# Set per-model options
llm 1min options set --model gpt-4o web_search true
llm 1min options set --model sonar num_of_site 10

# View all options
llm 1min options list

# View specific option
llm 1min options get web_search

# Remove an option
llm 1min options unset web_search

# Reset everything
llm 1min options reset

Priority hierarchy (highest to lowest):

1. CLI flags (-o web_search true)
2. Per-model config (--model gpt-4o)
3. Global defaults
4. Code defaults

List 1min.ai Models

# Show 1min.ai models with descriptions
llm 1min models

# Or see all models (including non-1min.ai)
llm models list | grep "1min.ai"

View Logs

# View last response
llm logs -n 1

# View last 5 responses
llm logs -n 5

Options Management

Manage persistent options that apply automatically to your models:

# Set global defaults
llm 1min options set web_search true
llm 1min options set num_of_site 5
llm 1min options set max_word 1000

# Set per-model options
llm 1min options set --model gpt-4o web_search true
llm 1min options set --model sonar num_of_site 10

# List all options
llm 1min options list

# List options for specific model
llm 1min options list --model gpt-4o

# Get specific option value
llm 1min options get web_search
llm 1min options get --model gpt-4o web_search

# Remove options
llm 1min options unset web_search
llm 1min options unset --model gpt-4o num_of_site

# Export/import configuration
llm 1min options export --output my-config.json
llm 1min options import my-config.json

# Reset all options
llm 1min options reset

Example config file (~/.config/llm-1min/config.json):

{
  "defaults": {
    "web_search": true,
    "num_of_site": 3
  },
  "models": {
    "gpt-4o": {
      "web_search": true,
      "num_of_site": 5
    },
    "sonar": {
      "num_of_site": 10
    }
  }
}

Conversation Management

The plugin provides commands to manage your 1min.ai conversations:

# List all active conversations
llm 1min conversations

# Clear conversation for a specific model
llm 1min clear --model gpt-4o-mini

# Clear all conversations
llm 1min clear --all

Advanced Conversation Management:

Use the included utility script for more options:

# List all conversations on 1min.ai server
python manage_conversations.py list

# Get details of a specific conversation
python manage_conversations.py get <conversation-uuid>

# Delete a specific conversation
python manage_conversations.py delete <conversation-uuid>

# Clear all conversations from server
python manage_conversations.py clear --all

# Export conversations to JSON
python manage_conversations.py export --output my-conversations.json

Features

• ✅ 75+ AI models from 10 providers through a single API
• ✅ Latest models: Claude 4.6 Sonnet/Opus, Claude 4.5 Haiku, GPT-5.4 / 5.1 Codex, o3 / o4 (incl. deep-research), Gemini 3.1, Qwen3 Max / Coder, DeepSeek V3.2, Grok 4, LLaMA 4, Magistral, Sonar
• ✅ Unified chat endpoint (/api/chat-with-ai) with structured settings
• ✅ SSE streaming: stream content chunks live via llm chat --stream
• ✅ Attachments: pass image keys / file IDs via images and files options
• ✅ AI memory: cross-conversation memory via with_memories
• ✅ Brand voice: per-call brand_voice_id
• ✅ Web search: real-time grounding on any model
• ✅ Mixed history: share conversation history across models (history_mixed)
• ✅ Persistent options: global and per-model defaults; options migrate for upgrades
• ✅ Test suite: 144 tests, 55% coverage
• ✅ CI/CD: GitHub Actions for automated testing on Python 3.8-3.12
• ✅ Conversation history tracking and management
• ✅ Specialized code generation mode
• ✅ Automatic conversation management
• ✅ List, export, and clear conversations
• ✅ Secure API key management
• ✅ Proper error handling with helpful messages
• ✅ Comprehensive model documentation

How It Works

This plugin uses the 1min.ai API v2 (/api/chat-with-ai) for chat models and the legacy /api/features endpoint for code generation:

1. Conversation Creation: When you start a chat, the plugin creates a conversation of type UNIFY_CHAT_WITH_AI on 1min.ai.
2. Chat Messages: Chat prompts POST to /api/chat-with-ai with structured promptObject.settings (web search, history, memory) and optional attachments (images, files).
3. Code Generation: When conversation_type=CODE_GENERATOR, prompts go to the legacy /api/features endpoint with the flat promptObject shape.
4. Streaming: Pass --stream (llm chat --stream ...) to receive server-sent events (content chunks streamed live; done terminates).
5. Context Management: Conversations are tracked per model in ~/.config/llm-1min/conversations.json.

API Endpoints Used

• POST https://api.1min.ai/api/chat-with-ai — chat (UNIFY_CHAT_WITH_AI), add ?isStreaming=true for SSE streaming
• POST https://api.1min.ai/api/features — CODE_GENERATOR only
• POST https://api.1min.ai/api/conversations — create conversation context
• GET https://api.1min.ai/api/conversations — list all conversations
• GET https://api.1min.ai/api/conversations/{uuid} — get specific conversation
• DELETE https://api.1min.ai/api/conversations/{uuid} — delete/clear a conversation
• POST https://api.1min.ai/api/assets — upload images/files (use the asset key returned in responses with the images / files options)

Authentication

The plugin uses the API-KEY header format (not OAuth/Bearer tokens).

Security Best Practices

API Key Management

• ✅ Never hardcode API keys in scripts or code
• ✅ Use environment variables: export ONEMIN_API_KEY="your-key"
• ✅ Or use LLM's secure key storage: llm keys set 1min
• ✅ Add .env files to .gitignore if used
• ✅ Rotate keys periodically

Script Security

All scripts in this repository follow security best practices:

• API keys are read from environment variables only
• No credentials are logged or printed
• Proper timeout values on all HTTP requests
• Error handling prevents information leakage
• Input validation where applicable

Development

Project Structure

llm-1min/
├── llm_1min.py              # Main plugin implementation
├── manage_conversations.py  # Conversation management utility
├── test_api.py              # API testing utility
├── tests/                   # Test suite (144 tests)
│   ├── test_options_config.py    # Options + legacy-key rejection (28 tests)
│   ├── test_model_execution.py   # Model execution + payload shape (18 tests)
│   ├── test_cli_commands.py      # CLI command tests (30 tests)
│   ├── test_streaming.py         # SSE parser tests (4 tests)
│   ├── test_attachments.py       # Attachments / memory / brand voice (8 tests)
│   ├── test_web_search_config.py # Nested settings.webSearchSettings (4 tests)
│   ├── test_web_search_debug.py  # Debug payload inspection (3 tests)
│   ├── test_debug_option.py      # Debug option behavior (3 tests)
│   ├── test_integration.py       # Integration + edge cases (42 tests)
│   ├── conftest.py               # Shared fixtures
│   └── fixtures/                 # Mock API responses
├── .github/workflows/       # CI/CD pipelines (144 tests passing)
│   ├── test.yml            # Automated testing (Python 3.8-3.12)
│   └── lint.yml            # Code quality checks
├── pyproject.toml          # Package configuration
├── README.md               # Main documentation
├── MODEL_SELECTION.md      # Comprehensive model guide
├── TESTING.md              # Testing documentation
├── CHANGELOG.md            # Version history
├── LICENSE                 # Apache 2.0 license
└── .gitignore             # Git ignore rules

Running Tests

This project includes a comprehensive test suite with 144 unit tests covering all major functionality:

# Install test dependencies
pip install -e .[test]

# Run all tests
pytest tests/ -v

# Run with coverage report
pytest tests/ -v --cov=llm_1min --cov-report=term-missing

# Run specific test file
pytest tests/test_options_config.py -v

# Run specific test
pytest tests/test_options_config.py::TestOptionsConfigSetters::test_set_option_global -v

Test Coverage:

• ✅ 144/144 tests passing (100%)
• ✅ 55% code coverage
• ✅ Options configuration + legacy-key rejection (28 tests)
• ✅ Model execution + payload shape (18 tests)
• ✅ CLI commands (30 tests)
• ✅ SSE streaming parser (4 tests)
• ✅ Attachments / memory / brand voice (8 tests)
• ✅ Web search config + debug (7 tests)
• ✅ Debug option behavior (3 tests)
• ✅ Integration and edge cases (42 tests)

See TESTING.md for complete testing documentation.

Testing the Plugin

1. Install in development mode:

   cd llm-1min
   llm install -e .

2. Verify installation:

   llm models list | grep 1min

3. Test with a simple prompt:

   llm -m 1min/gpt-4o-mini "Hello, world!"

4. Check logs for debugging:

   llm logs -n 1

Testing the API Directly

Use the included test script to verify API connectivity:

export ONEMIN_API_KEY="your-api-key"
python test_api.py

Comparison with Other Providers

Why Use 1min.ai?

Advantages:

• 🎯 Multi-model access: One API key for 75+ models across 10 providers
• 💰 Cost-effective: Often offers lifetime subscription deals
• 🔄 Model flexibility: Switch between OpenAI, Anthropic, Google, xAI, Mistral, Meta, etc.
• 🚀 Latest models: Quick access to Claude 4.6, GPT-5.4 / 5.1 Codex, o3 / o4, Gemini 3.1, Qwen3, DeepSeek V3.2, Grok 4, LLaMA 4
• 🌐 Diverse capabilities: From fast responses to complex reasoning to web-aware answers

Use Cases:

• Comparing responses from different models and providers
• Cost optimization by choosing the right model per task
• Avoiding vendor lock-in with multi-provider access
• Testing different reasoning approaches (O4, DeepSeek R1, Grok 4)
• Accessing models not available through direct APIs (Grok, Sonar)

Troubleshooting

"Authentication failed" Error

• Verify your API key is correct: llm keys set 1min
• Check that your 1min.ai account is active

"Rate limit exceeded" Error

• Wait a few moments before retrying
• Check your 1min.ai usage limits

No Response or Timeout

• Some models (like O3, O1) may take longer for complex reasoning
• Try increasing timeout or using a faster model

Model Not Found

• Ensure you're using the exact model ID from the Available Models list
• Check 1min.ai documentation for model availability

Debugging API Requests

To see exactly what's being sent to the API:

# Use debug option
llm -m 1min/gpt-4o -o debug true "your prompt"

# Or use environment variable
LLM_1MIN_DEBUG=1 llm -m 1min/gpt-4o "your prompt"

This will show:

• Options loaded from config files
• Options passed via CLI
• Final merged options
• Complete API payload being sent

Useful for troubleshooting:

• Why web_search isn't working as expected
• Which options are being applied
• Configuration conflicts between global and per-model settings

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

Apache 2.0

Related Projects

• omi — Standalone Go CLI sharing the same routing and API key. Single static binary, no Python runtime, sub-20ms cold start. Drop-in replacement when you don't need the LLM ecosystem.

Resources

• LLM Documentation
• 1min.ai
• Plugin Development Tutorial

Credits

Based on the LLM plugin architecture by Simon Willison.

Powered by 1min.ai - Multi-model AI API platform.