CATEGORY INTELLIGENCE / WORLDWIDE
Local LLMs.
Models and inference that run on your own machine.
Established red ocean
A crowded category without sustained search growth. Enter only with a concrete switching advantage.
Report dated · Method 1.0.2 · high evidence confidence.
- Matching active GitHub projects
- 847
- Search-interest growth
- -73% · last 8 complete weeks versus the previous 8
- Search term and region
- local llm · Worldwide
- Complete weekly observations
- 104
Why this classification
- Median weekly search interest fell 73% across two consecutive eight-week windows.
- 847 matching active repositories; the published dense-supply threshold is 50.
Source evidence
GitHub repository search · Collected 2026-09-15T16:35:44.775Z
topic:local-llm fork:false archived:false stars:>=5 pushed:>=2026-03-19
Google Trends search interest · Collected 2026-09-15T16:26:01.539Z
Leading repositories
| Repository | Stars | Description |
|---|---|---|
| HKUDS/nanobot | 48,187 | Ultra-lightweight, open-source, self-hosted personal AI agent framework in Python with WebUI, tools, memory, MCP, multi-agent workflows, automation, and chat apps |
| mozilla-ai/llamafile | 25,970 | Distribute and run LLMs with a single file. |
| LearningCircuit/local-deep-research | 9,094 | ~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted. |
| open-multi-agent/open-multi-agent | 6,925 | Self-hosted TypeScript agent runtime with durable approvals and verifiable run records. Own it, approve it, audit it. |
| MakazhanAlpamys/Soup | 6,584 | Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU. |
| dograh-hq/dograh | 5,654 | Open source voice AI platform. Self-hosted alternative to Vapi and Retell. On Prem, BYOK across Speech to Speech or LLM/STT/TTS, with a visual workflow builder, MCP native and telephony support. |
| maziyarpanahi/openmed | 5,324 | Local-first healthcare AI: clinical NER & HIPAA PII de-identification that runs 100% on-device. 2,200+ medical models, 21 languages, Apple MLX + Python, no cloud, no patient data leaving your network. Apache-2.0 |
| vinta/pangu.js | 4,822 | Opinionated paranoid text spacing in JavaScript, with on-device AI semantic judgment |
| langroid/langroid | 4,104 | Harness LLMs with Multi-Agent Programming |
| raullenchai/Rapid-MLX | 3,751 | The fastest local AI engine for Apple Silicon. 4.2x faster than Ollama, 0.08s cached TTFT, 100% tool calling. 17 tool parsers, prompt cache, reasoning separation, cloud routing. Drop-in OpenAI replacement. Works with Claude Code, Cursor, Aider. |
Limits of this result
- Google Trends measures relative search attention, not customers, revenue or willingness to pay.
- Supply counts active repositories carrying the selected GitHub topic; unlabeled and closed-source competitors are outside this coverage.
Search interest measures attention, not paying customers. Classification thresholds are published heuristics and still require empirical calibration.
Use and share the evidence
Permanent report · Markdown · JSON · PNG card
