Open Slap open-sources local agent engine

The pitch lands because it treats “local-first” as an architecture choice, not a marketing label. If the implementation holds up, this is the kind of project that can become a serious foundation for privacy-sensitive agent workflows.

• –The hybrid memory design is practical: exact-match FTS5 for names and APIs, vector search for semantics, and recent-message recall for short-term context.
• –Multi-provider support, including local and free-tier models, reduces lock-in and makes it easier to route tasks to the cheapest viable backend.
• –Permissioned automation is the right guardrail for a desktop agent that can touch files, web retrieval, and connectors.
• –MCP marketplace support broadens the system from “one assistant” into an extensible agent platform.
• –The repo is still early, but the stack choices are coherent: FastAPI, React/Vite, SQLite, and MoE routing make the project legible to builders.