Get started

localmelo documentation

localmelo is a pre-alpha local-first agent runtime with explicit memory layers, checked tool execution, and connectors for user-managed local or cloud model endpoints.

online loop verified memory active sleep planned

Set up

Install the package, configure a chat backend, then run a single direct query. This is the smallest useful path through the runtime.

minimum local run

pip install -e ".[dev,gateway]"
melo --reconfigure
melo "What is 6*7?"

What localmelo provides

Direct CLI Run one agent query from the command line and exit. Gateway API Serve the same runtime over HTTP with session reuse. Checked tool use Validate ingress, planning, execution, output, and memory writes. Layered memory Keep working memory, Hippo-routed long/profile/tools/personalized memory, and raw history separate. Backend connectors Connect to Ollama, MLC, vLLM, SGLang, or supported cloud providers. Roadmap tracking Follow runtime, memory, backend, and release-quality progress.

Current project state

Ready to verify

Online core loop

Direct CLI, gateway entry points, provider registry, tool execution, and checker boundaries have automated coverage.

Active

Memory persistence

SQLite-backed history and long-term memory are async and persistent. Track 2 v1 local-backend retrieval evidence is archived; retrieval policy is still evolving.

Later track

Sleep pipeline

Sleep-time training and evaluation are scaffolded. The personalization workflow is not part of the current baseline.