Persistent, cross-session memory for AI agents. Not context windows — real memory that survives between calls, users, and days.
5–10K tokens in GPU VRAM. Instant recall within a session. $0.003 per 1K tokens on RecordorAI hardware — 800x cheaper than GPT-5.4 for the same context density.
Full conversation history in pgvector + Supabase. Infinite retention, queried on demand. No per-token cost scaling with history length.
Each entity (user, project, decision) compressed to ~500 tokens. No hallucination risk — extractive, not generative. Costs almost nothing to store.
Auto-select Gemma 4 vs GLM-5.1 based on task complexity and cost. Fast tasks on Gemma, deep reasoning on GLM. No thought required.
Python and JavaScript. Works with LangChain, AutoGPT, CrewAI, or your own stack. Drop-in memory backend — no rewrites.
Browser-based UI for end users, connected to the same memory layer. Every conversation benefits from tiered memory automatically.
Supabase is SOC 2 Type II certified and HIPAA-eligible. Customer data stays isolated. Works for security-sensitive workloads.
vLLM, pgvector, Supabase — no vendor lock-in. The memory layer is yours. Deploy anywhere.
Plays well with the rest of the A.L.I.C.E. ecosystem — Runtime, Agents, Hub. Or use RecordorAI standalone with LangChain, CrewAI, AutoGPT, or your existing stack. Your choice.
Memory events stream in real time. Build reactive workflows on top of your agent's memory.
No API markups. No per-token royalties. You pay RecordorAI — we run the infra.
Free during beta. No credit card. Just access.
RecordorAI is the memory layer of A.L.I.C.E. — one of four products in a full-stack ecosystem. Each product is independently sold; each strengthens the others.