Local AI workflow
Local agent stack
An agent harness with a locally served model behind it — autonomy without an API bill. The stack: Hermes Agent, Goose, OpenClaw with Ollama serving 3 recommended local models — hardware readiness graded by the same engine as the feasibility checker.
Tools in this category most commonly run against cloud APIs; a local model served by the runtime below is the private, offline option this page sizes.
The stack
Tool
Hermes Agent
Alternative tool
Goose
Alternative tool
OpenClaw
Runtime
Ollama
Model pick
gpt-oss-20b
widely used open-weights pick for local agent loops
Model pick
qwen3-32b
stronger tool-following when 24GB-class memory is available
Model pick
llama-3.3-70b-instruct
the ambitious pick — needs big-memory hardware
Hardware readiness
Each cell is the engine's verdict for the best clean quant of that model on Ollama at 16K context — the same math as the checker, tiers not numbers.
Different hardware? Check your exact machine against any model in this stack.
Check your machine →Don't have the hardware? Build a stack for your budget →Template picks are editorial; every verdict, grade and tier above comes from the engine over sourced data — how the math works.