Run an uncensored AI agent locally
To run an uncensored AI agent locally, serve an abliterated open-weight model behind an OpenAI-compatible server on your machine and point the framework at it. Qwen 3.8 27B Abliterated fits one 24 GB GPU or a 32 GB Mac at 4-bit.
Models
| Model | Parameters | Size at 4-bit | Hardware | License |
|---|---|---|---|---|
| Qwen 3.8 27B Abliterated | 27.8B | about 17 GB | one 24 GB GPU or a 32 GB Mac | Apache 2.0 |
| Qwen 3.6 35B-A3B Abliterated | 36B total, 3B active | about 22 GB | one 24 GB GPU | Apache 2.0 |
| Gemma 4 31B Heretic | 31.3B | about 19 GB | one 24 GB GPU | Apache 2.0 |
| Llama 3.1 8B Abliterated | 8B | about 5 GB | an 8 GB GPU or a laptop | Llama 3.1 Community License |
Sizes at 4-bit are estimates. Models read on Hugging Face November 18, 2026.
Framework settings
| Framework | Local setting | Source |
|---|---|---|
| browser-use | ChatOpenAI(model, base_url="http://localhost:11434/v1", api_key="ollama") | README |
| Hermes Agent | base_url: http://localhost:11434/v1 | Providers |
| Goose | Built-in Ollama and LM Studio providers | Providers |
| Letta | /connect, then Ollama or LM Studio | Models |
| smolagents | OpenAIModel(api_base="http://localhost:11434/v1") | Models |
Local or hosted
| Local | Hosted | |
|---|---|---|
| Parameters | 8B to 36B | 320B to 2,780B |
| Context | Limited by your memory | 1,048,576 tokens |
| Hardware | A local GPU or Mac | None |
| Cost | Hardware and power | From $0.24 per 1M input tokens |
Hosted figures are for the models on this site.