Memory servers should be inspectable systems.

Ferrosa Memory gives AI agents typed recall: facts with history, plans, skills, graph edges, context folds, and auditable forgetting on a self-hosted MCP service.

Install locally Read the architecture

Fmmemory core

Forgeworkflows

Loomagent traces

Ferrosadurability

Atomstyped facts

A memory server, not a memory vibe.

Most agent memory products stop at text chunks and vector similarity. Ferrosa Memory treats memory as operational data that can be linked, queried, corrected, explained, and removed.

fact

Bi-temporal knowledge

Facts keep valid-time and transaction-time history, so agents can ask what is true now and what changed earlier.

graph

Entity relationships

People, projects, skills, sessions, documents, and tools are connected with typed edges instead of buried in prose.

recall

Hybrid retrieval

Recall blends lexical, vector, phonetic, graph, workspace, recency, and feedback signals before returning evidence.

forget

Auditable forgetting

Preview the blast radius, retract reversibly, or hard-delete with an append-only journal for trust and compliance work.

mcp

Agent-native tools

Expose a focused default toolset for daily work, with advanced graph, fold, Datalog, and operator tools when needed.

One system for every kind of agent memory.

Ferrosa Memory unifies facts, plans, traces, relationships, and context under a single durable service.

Forge

Build durable workflows.

Forge is the action layer: task plans, operator approvals, and automation paths that should survive beyond a single chat window.

Loom

Weave traces into recall.

Loom frames sessions, documents, and agent trajectories as threads that can be folded without losing provenance.

Ferrosa

Keep the substrate honest.

The database and operations layer: S3-backed durability, indexes, graph storage, and infrastructure you can run.

Memory

Make recall atomic.

Typed memory units, structured recall paths, and inspectable relationships between every fact, plan, and session.

MCPServer-first integration for Claude Code, IDEs, and agent runtimes.

21Focused everyday tools for ingest, recall, feedback, config, tasks, and forgetting.

58Advanced operator tools for folds, Datalog, graph inference, and batch work.

S3Durable storage path when you graduate from local development to a cluster.

Inspect the graph, then ship the memory server.

Ferrosa Memory should feel like a real systems product: workbench, graph visualization, operations view, local setup, and clear comparison pages.

Memory visualization. Follow entities, facts, and relationships instead of scanning opaque chunks.

Operations surface. Keep the database substrate visible when memory moves from laptop to cluster.

Start with one command.

Run local first, point your MCP client at the server, then decide whether your team needs a shared host or a Ferrosa-backed cluster.

Open quickstart Compare memory layers

$ curl -fsSL https://ferrosa-memory.com/setup-memory.sh | bash

$ ferrosa-memory serve
listening on 127.0.0.1:18765
mcp tools ready