Pre-installed Models

Docket comes with 12 AI models ready to use out of the box. Each model has a friendly name and is rated across three categories — Compute, Intelligence, and Safety — on a scale of 1 to 4.

Model Overview

#	Name	Size	Best For	Compute	Intelligence	Safety
01	Micro	0.6B	Runs on any hardware	★	★	★★★★
02	Spark	1B	Snappy and efficient	★	★★	★★★★
03	Lightning	1B	Ultra-fast for simple tasks	★	★	★★★★
04	Snapshot	4B	See and discuss images	★★	★★★	★★★★
05	Quick Chat	2B	Everyday conversations	★	★★	★★★★
06	Unchained	4B	Compact, no filters	★★	★★★	★
07	Storyteller	7B	Creative writing & roleplay	★★★	★★★	★★★
08	Unrestricted	7B	No content filters	★★★	★★★	★
09	Vision	7B	Understands images	★★★	★★★	★★★★
10	Developer	7B	Specialized for coding	★★★	★★★	★★★★
11	Balanced	9B	Good all-around performance	★★★★	★★★★	★★★★
12	Powerhouse	14B	Most capable	★★★★	★★★★	★★★★

Rating Key

• Compute — How much processing power the model requires (1 = lightweight, 4 = heavy)
• Intelligence — Quality of reasoning, knowledge, and response depth (1 = basic, 4 = advanced)
• Safety — How closely the model follows content guardrails (1 = unrestricted, 4 = strict safety filters)

Model Details

01 — Micro (0.6B)

Best for: Running on any hardware

• Smallest pre-installed model
• Minimal memory requirements
• Good for simple questions on low-spec machines
• May struggle with complex reasoning

02 — Spark (1B) 🟣 Vision

Best for: Snappy and efficient responses with image understanding

• Fast and lightweight
• Supports image input — can analyze photos and visual content
• Good balance of speed and vision capabilities

03 — Lightning (1B)

Best for: Ultra-fast simple tasks

• Fastest response times
• Great for quick answers and simple tasks
• Low resource usage

04 — Snapshot (4B) 🟣 Vision

Best for: Seeing and discussing images

• Dedicated image understanding model
• Describe photos, read text from images, answer visual questions
• More capable than Spark for image tasks

Usage:

1. Load the model
2. Click the attachment icon in chat
3. Upload an image
4. Ask questions about the image

05 — Quick Chat (2B)

Best for: Everyday conversations

• Natural conversational responses
• Good for general questions and chat
• Light on resources

06 — Unchained (4B) 🔴 Uncensored

Best for: Compact model with no content filters

• Fewer content restrictions in a smaller package
• Honest and direct responses
• Lower resource requirements than larger uncensored models

07 — Storyteller (7B)

Best for: Creative writing and roleplay

• Optimized for narrative and creative content
• Good at character voices and storytelling
• Handles longer form writing well

08 — Unrestricted (7B) 🔴 Uncensored

Best for: No content filters

• Fewer content restrictions
• Honest and direct responses
• Full 7B capability without safety guardrails

09 — Vision (7B) 🟣 Vision

Best for: Understanding images

• Most capable pre-installed vision model
• Can analyze complex images in detail
• Read text from images, describe scenes, answer visual questions

Usage:

1. Load the model
2. Click the attachment icon in chat
3. Upload an image
4. Ask questions about the image

10 — Developer (7B) 🔵 Coder

Best for: Programming assistance

• Optimized for code generation
• Understands many programming languages
• Good at debugging and explaining code
• Can write tests and documentation

Supported languages:

• Python, JavaScript, TypeScript
• Rust, Go, C/C++
• HTML, CSS, SQL
• And many more

11 — Balanced (9B)

Best for: Good all-around performance

• High quality responses across all tasks
• Strong reasoning capabilities
• Versatile — works well for conversation, analysis, and writing

12 — Powerhouse (14B)

Best for: Most capable pre-installed model

• Largest pre-installed model
• Best quality responses
• Handles complex instructions and reasoning
• Requires more RAM and processing time

tip

Start with smaller models and move up if you need more capability. Larger models don't always produce better results for simple tasks. Use the Compute, Intelligence, and Safety ratings to find the right model for your needs.

Real Model Names

Each friendly name maps to an underlying open-source model:

Docket Name	Underlying Model
Micro	SmolLM2 0.6B
Spark	Llama 3.2 1B
Lightning	Qwen 2.5 1B
Snapshot	Gemma 3N 4B
Quick Chat	Gemma 3N 2B
Unchained	Qwen 2.5 Uncensored 4B
Storyteller	Mistral 7B
Unrestricted	Qwen 2.5 Uncensored 7B
Vision	Qwen 2.5 VL 7B
Developer	Qwen 2.5 Coder 7B
Balanced	Gemma 2 9B
Powerhouse	Qwen 2.5 14B

Loading Models

1. Open the Models section in the sidebar
2. Click on a model to select it
3. Click "Load" to load it into memory
4. Wait for the status to show "Ready"

Unloading Models

Models stay in memory until you:

• Load a different model
• Close the application
• Manually unload via the Models page

Only one model can be loaded at a time for local models.