Releases·Tool Release
Developer releases golden-agent, a local Claude Code equivalent for local models
A Reddit user released golden-agent, a pip-installable coding agent that runs local models via llama.cpp as a Claude Code-style assistant. It bundles a tested three-tier model lineup sized to different VRAM budgets. It matters to practitioners wanting local coding agents without cloud APIs.
9 engagement1 sourceSun, Aug 30, 2026, 02:47 AM
AI Overview
- Install is two commands: pip install golden-agent, then golden-agent setup; models and the llama.cpp binary are lazy-downloaded
- Tiny tier uses LFM 2.5 2.6B (official QAT q4), needing 4GB VRAM or 8GB RAM
- Lite tier uses Ornith 1.5 9B (official Q4), needing 8GB VRAM or 12GB RAM
- Pro tier uses Qwen 3.8 27B (community QAT Q2), needing 16GB VRAM or 16GB RAM
- Author reports using Q5 KV cache across all tiers after extensive testing
Related
Social·Community
Users run Qwen 3.8 27B locally on consumer GPUs: 75 tok/s on 16GB, agentic coding on 12GB, AMD slowdowns
Local LLM users are sharing configurations for running Qwen 3.8 27B on consumer GPUs. An RTX 5080 setup reaches ~75 tok/s decode using a custom hybrid GGUF quant with multi-token prediction, while a 12GB laptop GPU runs it for agentic coding at 9–11 tok/s with Unsloth Dynamic 3.0 quants. An AMD RX 9060 XT user finds the 27B too slow and switches to Ornith 1.5 9B.
Local LLM users are sharing configurations for running Qwen 3.8 27B on consumer GPUs. An RTX 5080 setup reaches ~75 tok/s decode using a custom hybrid GGUF quant with multi-token prediction, while a 12GB laptop GPU runs it for agentic coding at 9–11 tok/s with Unsloth Dynamic 3.0 quants. An AMD RX 9060 XT user finds the 27B too slow and switches to Ornith 1.5 9B.
Updated Mon, Aug 31, 2026, 09:34 AM — An RTX 5080 user reports ~75 tok/s decode, sometimes 100+ tok/s with good MTP acceptance, on the jrell hybrid quant.
210952 engagement4 sourcesredditThu, Aug 20, 2026, 11:54 PM
Social·Community
Best Local Models for Claude Code on 8GB VRAM: Tested Picks from Gemma4 12B to Muse Glimmer 30B
A Reddit user shared results of running coding-capable local models with Claude Code on a laptop with an RTX 5060 (8GB VRAM) and 16GB RAM. The post lists five quantized Ollama models — Gemma4 12B, Qwen3.5 9B, Qwopos3.5 4B, Qwen3.5 4B, and Muse Glimmer 30B — with context-length notes for each. It serves as a practical reference for developers running agentic coding tools on consumer hardware.
A Reddit user shared results of running coding-capable local models with Claude Code on a laptop with an RTX 5060 (8GB VRAM) and 16GB RAM. The post lists five quantized Ollama models — Gemma4 12B, Qwen3.5 9B, Qwopos3.5 4B, Qwen3.5 4B, and Muse Glimmer 30B — with context-length notes for each. It serves as a practical reference for developers running agentic coding tools on consumer hardware.
26 engagement1 sourceredditFri, Sep 11, 2026, 04:49 PM
Social·Community
User seeks local LLM for AI agents with 8GB VRAM
A user with an RTX 5070 8GB and 64GB RAM asks for a local LLM that works with Qwen Code or OpenCode for small tasks. They tested Qwen variants but found them too slow or failing tool calls. The community may suggest models that fit in 8GB VRAM.
A user with an RTX 5070 8GB and 64GB RAM asks for a local LLM that works with Qwen Code or OpenCode for small tasks. They tested Qwen variants but found them too slow or failing tool calls. The community may suggest models that fit in 8GB VRAM.
30 engagement1 sourceredditTue, Aug 25, 2026, 11:21 PM
Products·launched
Local LLM agents for Claude Code via MCP
A setup that integrates local LLMs (e.g., Qwen3.8-27B) as agents in Claude Code using MCP, offloading well-specified tasks to local models to reduce token usage on Anthropic's paid models. It allows using Opus for main model and local models for sub-agents within the same session and context.
A setup that integrates local LLMs (e.g., Qwen3.8-27B) as agents in Claude Code using MCP, offloading well-specified tasks to local models to reduce token usage on Anthropic's paid models. It allows using Opus for main model and local models for sub-agents within the same session and context.
35 engagement1 sourceredditThu, Aug 27, 2026, 12:03 AM
Social·Community
Users compare Qwen3.6-35B-A3B and Qwen3.8-27B for local coding on RTX 5070 Ti
A developer on Reddit asks for advice choosing between two local LLMs for coding assistance on an RTX 5070 Ti (16GB VRAM) with 32GB RAM. They report that Qwen3.8-27B-UD-IQ3_XXS worked well for planning and task setup, and they are now testing Qwen3.6-35B-A3B-unsloth-nvfp4-fast-mtp for implementation.
A developer on Reddit asks for advice choosing between two local LLMs for coding assistance on an RTX 5070 Ti (16GB VRAM) with 32GB RAM. They report that Qwen3.8-27B-UD-IQ3_XXS worked well for planning and task setup, and they are now testing Qwen3.6-35B-A3B-unsloth-nvfp4-fast-mtp for implementation.
290 engagement2 sourcesredditMon, Aug 17, 2026, 04:11 PM