Open source · Apache 2.0 · Engine for AI agents

Memory. Journal. Ethics. For AI agents.

An engine, not an app. The memory, journaling and ethics substrate you embed inside other software — agents, assistants, tools, pipelines. Speaks MCP, so any compatible client gets persistent memory, a first-person journal, an auditable ethics engine and a per-user biological clock.

View on GitHub
[ Apache 2.0 in full · 61 MCP tools · hash-chained journal ]
scroll to explore
01 — What is Celiums

An engine,
not
an app.

Celiums Memory has no UI. No dashboard to log into. It is the substrate you embed inside other software — and it speaks MCP, so any compatible client (Claude Code, Cursor, Continue, Cline, OpenCode, or your own) gets persistent memory, a first-person journal, an auditable ethics engine and a per-user biological clock without you building any of it.

Every request flows through the same path: auto-bootstrap prepends the session context, RBAC + AAL + Ethics apply three orthogonal checks, and the dispatcher routes to one of 61 typed MCP tools. Storage adapter and LLM provider are config-only choices.

Apache 2.0 in full. No open-core split, no paid tier, no proprietary core held back. The Ethics Engine — every layer — is open and auditable. The component that makes moral calls is the one that least deserves to be hidden.

Quick start

Clone, compose up, you're running.

Docker brings the Postgres + Qdrant + Valkey stack online in one command. For embedded use, SQLite mode needs no external services at all.

$
git clone https://github.com/terrizoaguimor/celiums-memory.git && cd celiums-memory && pnpm docker:up
02 — The engine

Seven primitives,
one substrate.

Each primitive is its own typed MCP tool family. Compose them, ignore them, replace them — the engine doesn't enforce a workflow. It enforces guarantees.
03 — The pipeline

Four stages,
every call.

The same path runs for every MCP request and every HTTP call. Three orthogonal checks gate every irreversible op; the dispatcher only sees calls that have already passed.
04 — The Ethics Engine

Four layers.
Fully open.

The Ethics Engine is the one component that makes moral calls. It is also the one that least deserves to be hidden. Every layer is auditable, every decision is traced.
Read the architecture →
05 — Built for

MCP clients,
any stack.

Any MCP-compatible client points at the HTTP endpoint with an API key and immediately gets `remember`, `recall`, `journal_write`, `forage`, `ethics_trace` and the rest — with auto-bootstrap loading prior context into the first reply.
06 — Tool surface

Sixty-one
typed tools.

Every primitive the engine exposes is an MCP-typed tool. Clients invoke only what they need; the rest stays out of context. All families share the same RBAC + AAL + Ethics gate.
07 — Questions

Frequently
asked.

If something's not here, open an issue on GitHub — that's where we answer the rest.
08 — Embed the engine

Clone it.
Run it. Own it.

Apache 2.0 in full. Every layer of the engine — including the Ethics — is yours to inspect, fork, and run. No proprietary core, no waitlist, no telemetry you didn't consent to.

View on GitHub

[ Apache 2.0 · 61 MCP tools · 4-layer ethics · hash-chained journal ]