Engram is an LLM agent with persistent memory: every task becomes an episode, and during sleep it consolidates lessons into reusable skills. Over time it gets better at problems it has seen before — without retraining.
Free options first — Engram detects them automatically once configured.
ollama run qwen2.5:7b.console.groq.com.gemini-1.5-flash. Get a key at aistudio.google.com.Paid options: Anthropic Claude, OpenAI GPT, DeepSeek, Kimi, Qwen, Mistral, and more.
Open Chat, type a task in any language, press Send. The agent thinks, uses tools (Python sandbox, AST checks), retrieves any past skills, and answers. Every conversation creates an episode visible under Episodes. After 3+ episodes, run a sleep cycle to consolidate skills — they appear under Skills with their genealogy under Lineage.