Show HN: LLM-use – Open-source tool to route and orchestrate multi-LLM tasks

I built llm‑use, an open‑source Python framework for orchestrating large language model workflows across local and cloud models with smart routing, cost tracking, session logs, optional web scraping, and optional MCP integration. It’s designed for agent workflows (planner + workers + synthesis) that leverage multiple LLMs without manual switching or custom glue code.

Examples

Simple local usage:

ollama pull llama3.1:70b
ollama pull llama3.1:8b

python3 cli.py exec \
–orchestrator ollama:llama3.1:70b \
–worker ollama:llama3.1:8b \
–task “Summarize 10 news articles”

This runs a planner + worker flow fully locally.

Hybrid cloud + local usage:

export ANTHROPIC_API_KEY=”sk-ant‑…”
ollama pull llama3.1:8b

python3 cli.py exec \
–orchestrator anthropic:claude-3-7-sonnet-20250219 \
–worker ollama:llama3.1:8b \
–task “Compare 5 products”

export ANTHROPIC_API_KEY=”sk-ant‑…”
ollama pull llama3.1:8b

python3 cli.py exec \
–orchestrator anthropic:claude-3-7-sonnet-20250219 \
–worker ollama:llama3.1:8b \
–task “Compare 5 products”

Routes tasks between cloud provider models and a local worker.

TUI chat mode:

python3 cli.py chat \
–orchestrator anthropic:claude-3 \
–worker ollama:llama3.1:8b

Interactive CLI chat with live logs and cost breakdown.

Why it matters
• Orchestrate multiple LLMs — OpenAI, Anthropic, Ollama/llama.cpp — without writing custom routing logic.
• Smart routing and fallback — choose better models for each task and fall back heuristically or learned over time.
• Cost tracking & session logs — see costs per run and preserve history locally.
• Optional scraping + caching — enrich tasks with real web data if needed.
• Optional MCP server integration — serve llm‑use workflows via PolyMCP.

llm‑use makes it easier to build robust, multi‑model LLM systems without being tied to a single API or manual orchestration.

Repo: https://github.com/llm‑use/llm‑use