Sibyl Memory · Base mainnet · Virtuals compute

The agent that
cannot forget.

Most agents lose everything the moment the tab closes. Mnemos keeps memory durable, gates money on it, and refuses to act on a memory that turned out wrong.

mnemos
$ mnemos revise preference contractor_rate 60 --reason "renotiated"
revised preference:contractor_rate: 40 -> 60
  suspect: agreement:fencing
$ mnemos pay fencing 160
gate: agreement 'fencing' is suspect: depends on revised memory
      contractor_rate; reconsider first
payment refused
147tests, all green
2partner stacks, live
1revision gate
17MCP tools, hosted

A wrong memory
stops the money.

Every agent has the same correction story. You say the rate is wrong, the agent says noted, and the payment still goes out. Mnemos computes the blast radius of every correction, marks dependents suspect, and the gate refuses until each one is explicitly reconsidered.

1Correct

revise the fact, history stays

2Contaminate

dependents become suspect

3Close

the payment gate refuses

4Reconsider

review reopens the gate

"a correction that does not propagate is a new fact sitting next to an old decision. walking every agreement built on the wrong number is the part nobody builds because it is tedious. tedious is where the money leaks."

Sibyl Cap, on the revision gate

What memory improves

01

Money

Recall costs zero tokens. The same fact re-reads forever, and keepsakes restore a full agent with one command.

02

Decisions

The gate refused a real payment while the agreement was only agreed, then executed it once delivered. Memory alone changed the outcome.

03

Failures

A high severity lesson vetoes linked payments until resolved. Mistakes stop repeating.

04

Corrections

When a memory turns out wrong, everything built on it becomes suspect until reviewed. Nothing deleted, all journaled.

How it learns

01

Dream

Background consolidation over the journal. Repeated patterns, unanswered questions, and refused payments become review-gated proposals. Nothing applies itself.

02

Rewind

Memory at any past moment. Walk the revision history backwards and diff what changed: what did we believe, and when did it change.

03

Entities

Deterministic entity extraction on every fact. Recall boosts facts that share an entity with the question. Zero model calls.

04

Hybrid search

FTS first, semantic assist when you bring a key. An optional embedder fuses cosine with keyword search, off by default.

05

Pulse

A bounded proactive queue grounded in memory. The most urgent matter surfaces, one per tick, and declined matters stay declined.

06

Owner profile

A curated summary of who you are: identity, preferences, accepted principles, standing agreements. Not a raw history dump.

Measured, not marketed

Seeded, reproducible runs of the real gate. 12 trials, seed 1337, no fabricated numbers.

ArmAllowedRefused
Memory on: remembered, delivered agreement120
Memory off: same request, wiped store012
Revision on: suspect until reconsider012
Revision off: same agreement, no suspect check120

Reproduce with scripts/ablation.py. Pinned by tests.

Full captures and reproduction for both partner stacks: docs/VERIFICATION.md.

Use on your agent

One config line. Your agent gets durable memory, the payment gate, and revision.

Claude Desktop

{
  "mcpServers": {
    "mnemos": {
      "url": "https://mnemos-production-2572.up.railway.app/mcp"
    }
  }
}

Runs on the hosted endpoint. No install, works on any machine.

VS Code / Cursor

{
  "servers": {
    "mnemos": {
      "type": "http",
      "url": "https://mnemos-production-2572.up.railway.app/mcp"
    }
  }
}

Streamable HTTP, works straight from the mcp.json file.

Remote clients

https://mnemos-production-2572.up.railway.app/mcp

Works with ChatGPT connectors, Claude web, and any streamable HTTP client. One click install on Smithery.

Talk to the live agent

Real memory, real tools, and an agent model answering. Chips run raw tools, the box is a conversation.

connecting

Private memory: each browser gets its own store, and reopening keeps it.

MNEMOS
Hi, I am Mnemos. My memory is empty right now, so to me you are a stranger until you tell me something. Teach me a fact and I will keep it.

Give your agent
a memory it cannot lose.

Connect the MCP endpoint and Mnemos remembers, gates, and corrects for you.