Memorall

Memorall — The browser is your agent's harness. Memorall turns your browser into a full agent workspace — with memory, tools, files, and model choice all in one place. Capture context from any page, build a knowledge graph from your real work, and run a powerful local-first agent that can browse, read documents, execute code, and remember everything across sessions. No fragmented apps, no mandatory cloud, no context lost between tabs. Open Source: https://github.com/zrg-team/memorall ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🧠 MEMORY — Durable memory that learns from your browser Pages, selections, screenshots, PDFs, notes, and spreadsheets get captured from any tab and converted into structured, topic-scoped memory. Relationships between ideas, sources, and documents stay attached, so the work stays recoverable instead of disappearing after one session. A built-in knowledge graph connects facts across everything you save. • Capture from pages, selections, screenshots, files • Topic-organized memory with a live knowledge graph • Hybrid retrieval: exact match + semantic search • Sources and relationships stay linked to prior work ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ⚡ AGENT POWER — One strong agent, fully inside your browser The agent is not isolated. It has live browser access, virtual filesystems, your personal documents, and a browser-hosted sandbox runtime — all from the same product. • Browser session — live page content, DOM awareness, page-aware actions • Workspace files — virtual filesystem + document library (PDF, MD, Excel, and more) • Node.js sandbox — run code, install packages, start APIs, prototype web apps locally • Visual flow builder for custom agent behavior and logic • MCP-ready architecture for future extensibility ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🏠 OFFLINE-FIRST — Your stack, your control The core experience works without any external service. No account required to get started. • All data stored locally via PGlite — no server, no sign-in required • Local memory, local files, local agent — fully operational offline • No mandatory cloud dependency for core workflows • Optional Supabase auth available when you want sync ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ⚙️ MODEL CHOICE — Browser-local LLMs are first-class, not fallbacks Use whichever runtime fits the job. The agent workflow stays the same regardless of where inference runs. • Browser-local inference: WebLLM, Wllama, Transformers.js • Local server runtimes • Remote APIs when you choose them: OpenAI, OpenRouter • Switch providers without changing the workflow ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 💬 SUPPORT • GitHub: https://github.com/zrg-team/memorall • Issues: https://github.com/zrg-team/memorall/issues