Continuum

Continuum tackles the context amnesia problem in AI coding sessions by giving all your MCP clients a shared memory layer backed by SQLite and sqlite-vec. It exposes tools like `knowledge_create`, `knowledge_search`, and `knowledge_get` for storing and retrieving semantic lessons, plus `agent_dispatch` and `registry_list` for orchestrating isolated worktree sessions with collision detection. The setup seeds every project with a PLOT.md protocol that walks through Intake, Research, Verify, and Handoff phases. One client reads the plot and becomes the orchestrator, then hands off to fresh agents that can query what earlier sessions learned. The default embedder runs in-process with Transformers.js, or you can point it at Ollama or any OpenAI-shaped endpoint. Everything lives in one local SQLite file with no SaaS dependencies.