Get the latest setups
THE HANDS-ON STARTING POINT

What would you like to run?

Choose a goal. Pick your machine or rent a GPU. Follow one setup all the way to a working result.

Choose a task Learn the concepts
Real workflows Hardware before commands A first-run check
01 Choose a task02 Choose hardware03 Follow a setup04 Verify the result
THE VISUAL BEGINNER’S GUIDEWant to learn the basics of local and uncensored AI?Understand models, hardware, runtimes, privacy, and what “uncensored” really means.Explore the guide
01 / START WITH THE OUTCOME

Choose what you want to do.

Pick a task to see a recommendation and every relevant model already in the catalog.

02 / CHOOSE YOUR COMPUTE

Where will you run it?

Your task stays the same. The best model and exact package may change with your hardware.

Recommendations use task relevance and source evidence. Save a machine to see fit ratings. Save my machine

03 / A RECOMMENDATION YOU CAN CHECK

Private AI agents, from start to result.

Recommendations draw from the published catalog and editor-approved shortlist. Open the exact workflow for commands and files.

RECOMMENDED MODEL Bonsai 2 · 27BMAIN APP PrismML llama.cpp fork + Hermes AgentFULL GUIDE Build a local tool-using agent with Bonsai 2 · 27B
Freshness 90 out of 100. Release age: 28 days old, Sep 17, 2026. Measures age, not quality.
Popularity 90 out of 100. 4,849,199 downloads in 30 days; 2,697 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.
What do these scores mean?

Two ways to compare models. Neither score measures answer quality, benchmark performance, or whether a model fits your machine.

Freshness · release age

A new release starts at 100. The score halves every 180 days: 100 × 2^(−age in days / 180). When the release date is unavailable, a repository date is a fallback—not proof of when the model was released. Missing dates stay unscored.

Check the date source
Popularity · adoption

Downloads over the past 30 days carry 80% of the score; lifetime likes carry 20%. Both use a logarithmic scale, capped at 10 million downloads and 100,000 likes. If likes are unavailable, downloads carry all the weight. Missing download statistics stay unscored.

round(80 × min(1, log10(1 + downloads) / 7) + 20 × min(1, log10(1 + likes) / 5))

Check the statistics source
Scores as of Oct 15, 2026. Statistics checked Oct 15, 2026. Statistics are marked for refresh after 72 hours.Full scoring method

Why this is hereAn editor-approved pick with a published task-specific workflow.

01

Check the exact package

The guide lists the supported files, software, and hardware for this model.

Compare setup options
02

Follow the supported route

Choose local hardware or a rented GPU in the full guide.

Review the setup
03

Verify a real result

Hermes connects to the local OpenAI-compatible endpoint, the model returns a tool call, the client sends the tool result back, and the final answer reflects that result. Check the llama-server log for the tool-call exchange.

Open the first-run check
Open the complete setup guide
EXPLORE THE WHOLE TASK

All models for private ai agents.

61 of 61 published models match these filters for this task. New catalog entries appear here automatically; featured picks require editorial approval.

These filters apply to the list below. Models missing the required source data are excluded when a filter is active.

IN THE CATALOG

Kolibri 1 · BF16

An open-weight English/German reasoning and tool-calling model with a publisher-maintained vLLM plugin and multi-GPU serving instructions.

Freshness 95 out of 100. Release age: 12 days old, Oct 3, 2026. Measures age, not quality.
Popularity 45 out of 100. 2,254 downloads in 30 days; 60 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.
Build a private AI agent with Kolibri 1
IN THE CATALOG

GLM-5.3

Z.ai’s open-weight ~744B-total / ~40B-active reasoning model for long-running agents and coding. Its native FP8 checkpoint has a documented 8×H200 Linux vLLM path; a rented multi-GPU Pod is the practical alternative to owning that server.

Freshness 82 out of 100. Repository age: 51 days old, Aug 25, 2026. Measures age, not quality.
Popularity 85 out of 100. 1,803,969 downloads in 30 days; 2,223 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.
Arena Open Source Agents: rank 6 of 18. Ranked model: GLM 5.3 (Max). Export dated Oct 14, 2026. Source checked Oct 15, 2026. Human preference ranking, not hardware suitability.+2
Run GLM-5.3 with Strix for Authorized Security Reviews
EDITOR-APPROVED

GLM-5.3 Flash

Z.ai’s open-weight 321B-total / 18B-active multimodal agent model. Self-host the native FP8 checkpoint on a documented multi-GPU Linux server or rented GPU Pod; use Strix for an authorized repository review.

Freshness 82 out of 100. Repository age: 51 days old, Aug 25, 2026. Measures age, not quality.
Popularity 92 out of 100. 6,531,090 downloads in 30 days; 2,855 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.
Arena Open Source Agents: rank 9 of 18. Ranked model: GLM 5.3 Flash. Export dated Oct 14, 2026. Source checked Oct 15, 2026. Human preference ranking, not hardware suitability.+3
Build a private AI agent with GLM-5.3 Flash Run GLM-5.3 with Strix for Authorized Security Reviews
EDITOR-APPROVED

Bonsai 2 · 27B

A compressed 27B-class reasoning model with publisher-provided GGUF packs and a dedicated llama.cpp fork for CUDA, Metal, and CPU.

Freshness 90 out of 100. Release age: 28 days old, Sep 17, 2026. Measures age, not quality.
Popularity 90 out of 100. 4,849,199 downloads in 30 days; 2,697 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.
Build a local tool-using agent with Bonsai 2 · 27B
EDITOR-APPROVED

Qwen3.8-27B

A dense vision-language model for coding, reasoning, and agent tasks, with image and video understanding.

Freshness 76 out of 100. Repository age: 71 days old, Aug 5, 2026. Measures age, not quality.
Popularity 95 out of 100. 7,363,968 downloads in 30 days; 17,749 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.
Arena Open Source Web coding: rank 10 of 55. Ranked model: qwen3.8-27b. Export dated Oct 13, 2026. Source checked Oct 15, 2026. Human preference ranking, not hardware suitability.+3
Build a local AI agent with Qwen3.8 27B
IN THE CATALOG

RavenX CyberAgent · 35B-A3B

A security-focused, open-weight 35B mixture-of-experts model (about 3B active parameters) distributed as GGUF. Run it locally with llama.cpp on Windows, macOS, or Linux, or use a rented RunPod GPU.

Freshness 60 out of 100. Repository age: 133 days old, Jun 4, 2026. Measures age, not quality.
Popularity 58 out of 100. 18,756 downloads in 30 days; 189 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.
Run RavenX CyberAgent for Authorized AI Pen Testing
IN THE CATALOG

Qwen3-Coder-Next · 80B A3B

Qwen3-Coder-Next is an open-weight text-generation model from Qwen/Qwen3-Coder-Next, licensed under apache-2.0. It is designed specifically for coding agents and local development. The model has 80B total parameters with 3B activated parameters, a 262,144-token native context length, and a hybrid architecture combining Gated DeltaNet, Gated Attention, and Mixture-of-Experts layers. It supports only non-thinking mode and does not generate thinking blocks.

Freshness 37 out of 100. Repository age: 258 days old, Jan 30, 2026. Measures age, not quality.
Popularity 79 out of 100. 588,526 downloads in 30 days; 1,696 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.
Build a local AI agent with Qwen3-Coder-Next
IN THE CATALOG

LFM2.5 2.6B

LFM2.5-2.6B is a text-generation model from LiquidAI with 2.69B parameters, a 131,072-token context window, and a 128,000-token vocabulary. It uses a hybrid architecture with 30 layers, including 22 double-gated short convolution blocks and 8 GQA layers, and was trained on 34 trillion tokens. The model is described as general-purpose, text-only, and post-trained for agentic workloads, with multilingual coverage across 16 languages. It is distributed under the lfm1.0 license and supports safetensors and text-generation tags.

Freshness 74 out of 100. Repository age: 79 days old, Jul 28, 2026. Measures age, not quality.
Popularity 70 out of 100. 114,916 downloads in 30 days; 831 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.
LFM2.5 2.6B local agent setup
04 / KNOW THAT IT WORKS

Don’t stop at “installed.”
Make something.

A working setup produces an output you can inspect. Start with the guide’s small test, then increase quality, context, or duration once it is stable.

SUCCESS CHECK / PRIVATE AI AGENTS

Hermes connects to the local OpenAI-compatible endpoint, the model returns a tool call, the client sends the tool result back, and the final answer reflects that result. Check the llama-server log for the tool-call exchange.

If the first test fails, check the guide’s package, runtime, platform, and troubleshooting steps before changing to a larger setup.
READY FOR THE NEXT STEP?

Make the setup yours.

Save your hardware, compare other models for the same task, or learn the ideas behind your first run.

Add my hardware Browse all workflows Read the full beginner’s guide

Hardware fit is labeled by its evidence. Model and workflow pages contain the full sources, package requirements, and supported platform routes. RunPod links are affiliate links.