SIBYL LABSdocs

Sibyl Memory Plugin V3 live

The memory layer for AI agents. Lives on your machine, organizes information by what kind of thing it is, and recalls it the way a human would expect. Works with Claude Code, Codex, Cursor, Continue, Hermes, and any Python framework. Free to start, paid plans for power users. Live on PyPI.

From blank machine to working memory in ~5 minutes. Install · activate via wallet or terminal-pairing code · upgrade in USDC · cap lifts before you put your phone down.

What it does, in one sentence

Sibyl Memory Plugin gives your AI agent a long-term memory that persists across sessions, knows the difference between a fact and a conversation, and stays on your computer.

Start here

Install

One command to install. The full walk-through from blank machine to working memory.

● live · ~5 min

Quickstart

Write something to memory, read it back, search across everything. Your first session.

● live · ~10 min

How memory works

The five kinds of memory and why splitting them up makes the agent smarter.

● live · 7-min read

What's free, what's paid

The free tier is generous. Paid plans unlock self-learning, the memory check-up, and remove the size cap.

● live

What you get on the free tier

  • Five-tier memory model. Your agent can store conversations, facts, rules, lookups, and archived items separately, and recall each in the way that fits.
  • Fast search across everything. Type a phrase, find everywhere the agent has seen it.
  • Multiple identities on one machine. The agent can keep separate memory for work vs. side projects vs. team channels, all on the same install.
  • 2 MB local storage cap. Enough for a few weeks of real use. The agent warns you as you approach.
  • Activation and account on us. Sign in with your wallet, no credit card required.

What unlocks on the paid tier

  • No size cap. Use it as long as you want.
  • Self-learning. Your agent watches how you work and proposes reusable skills you can teach it with a click. How it works →
  • Memory check-up. A health report that surfaces duplicates, stale entries, and anything that looks off. How it works →
  • Cross-device sync (Sync plan, coming soon).
  • Shared team memory (Team plan, coming soon).

See the full plan comparison →

V3 · payment, auth, account new

The V3 release shipped a full payment + authentication + account-management stack. Pick up here once you've completed the install:

Authentication

SIWE wallet sign or 6-digit terminal pairing code. HMAC-signed credentials. Session-bound nonces, replay-protected.

● live · V3 · ~8 min

Payment & upgrade

USDC subscribe on Base, stake $SIBYL for token-aligned tier, or lifetime. Cross-account replay defense built in.

● live · V3 · ~10 min

Account & status

Three ways to check your tier, cap usage, and server health. The terminal is the dashboard.

● live · V3 · ~7 min

CLI reference

The sibyl command, in full. Four subcommands: init, upgrade, status, health.

● live · V3 · reference

Integrations

Claude Code, Codex, Cursor, Continue, Hermes, LangChain, LlamaIndex. MCP server, native provider, or direct SDK.

● live · V3 · ~9 min

What's live as of V3 (2026-05-16)

Production-validated

All three packages are live on PyPI: sibyl-memory-client v0.3.2, sibyl-memory-hermes v0.2.2, and the sibyl-memory-cli v0.1.0 that ships the sibyl command. The full payment rail is production-validated end-to-end (real USDC, real on-chain subscription, real tier flip). Hard cap is server-enforced. Activation works via wallet (SIWE) or 6-digit terminal pairing code. Credentials are HMAC-signed. A 7-agent audit completed and remediated 19+ Tier-0/1/2/3/4 findings before the release.