Seven primitives, one graph

Supermemory bundles what would otherwise be seven separate systems: memory, retrieval (SuperRAG), a POSIX filesystem mount, user profiles, source connectors, document extractors, and qualitative analysis. All backed by one custom graph engine at sub-300ms. Ships enterprise (on-prem, VPC, air-gapped) and consumer (personal second-brain app) from the same substrate.

Explore this profile

Seven primitives, one graph, one API

The core architectural bet: every primitive resolves into a single graph that your agent traverses at request time. Not seven databases stitched together with adapters, one graph with seven ways in.

                     SUPERMEMORY — 7 PRIMITIVES, ONE GRAPH

  ┌────────────────────────────────────────────────────────────────────────┐
  │  YOUR AGENT / APP / PERSONAL SECOND BRAIN                              │
  └───────────────────────────────────┬────────────────────────────────────┘
                                      │
                    supermemory API · POSIX filesystem mount
                                      │
                                      ▼
  ┌────────────────────────────────────────────────────────────────────────┐
  │  SUPERMEMORY  ─  seven ways in                                          │
  │                                                                        │
  │   01 MEMORY        continual learning, structured, graph-native         │
  │   02 SuperRAG      hybrid search + rerank, sub-300ms retrieval          │
  │   03 FILESYSTEMS   POSIX mount — grep=semantic, profile.md live         │
  │   04 PROFILES      per-peer identity/preference/behavior state          │
  │   05 CONNECTORS    Slack · Notion · Drive · Gmail · GitHub · S3         │
  │   06 EXTRACTORS    PDF · web · image · audio · smart chunking           │
  │   07 QUAL ANALYSIS cluster/summarise/explain in-place                   │
  │                                                                        │
  │                                      │                                 │
  │                                      ▼                                 │
  │                       ┌────────────────────────────┐                   │
  │                       │  UNIFIED GRAPH              │                   │
  │                       │  custom engine             │                   │
  │                       │  <300ms traversal          │                   │
  │                       │  dynamic dreaming (async)   │                   │
  │                       └─────────────┬──────────────┘                   │
  └─────────────────────────────────────┼──────────────────────────────────┘
                                        │
                                        ▼
  ┌────────────────────────────────────────────────────────────────────────┐
  │  YOUR DATA — pulled in automatically by connectors                     │
  │  Slack · Notion · Drive · Gmail · GitHub · S3 · custom sources         │
  └────────────────────────────────────────────────────────────────────────┘

The filesystem mount is genuinely unique

None of the other three products (Orectic, Penumbra, Honcho) has this. Supermemory mounts as a native POSIX filesystem on macOS and Linux. That means:

• grep becomes semantic search across everything the agent has ingested
• cat profile.md reads a live, synthesized profile that updates as new context arrives
• ls shows an entity-organized view of memory as directories
• Any agent that shells out (Claude Code, Codex, Cursor) gets memory access for free without needing an SDK integration

This is the piece that quietly changes what "agent tooling" means. Instead of "here's an SDK, learn our API," it's "the memory is already at the paths your agent already reads from."

Dynamic dreaming

Same terminology Honcho uses for its background async reasoning. Supermemory launched it as the default in a recent release. Two independent teams landing on the same primitive is a category signal worth taking seriously — see the head-to-head section for the implications.

Where Supermemory sits in the 4-way

The other three products each picked a lane. Honcho went narrow-and-deep on peer memory. Orectic and Penumbra bundled a few primitives around one bet (typed business substrate). Supermemory bundled everything. It's the "why pick" play.

                     SCOPE — HOW MUCH EACH PRODUCT SHIPS

  narrow                                                              broad
  ──────                                                              ─────

  ┌────────────┐    ┌────────────┐    ┌────────────┐    ┌──────────────────┐
  │ HONCHO     │    │ ORECTIC    │    │ PENUMBRA   │    │ SUPERMEMORY      │
  │            │    │            │    │            │    │                  │
  │ peer mem + │    │ files →    │    │ ontology → │    │ memory + RAG +   │
  │ reasoning  │    │ typed KG + │    │ typed subs+│    │ connectors +     │
  │ per query  │    │ Oracle     │    │ your agent │    │ filesystem +     │
  │            │    │ agent      │    │ substrate  │    │ profiles +       │
  │            │    │            │    │            │    │ extractors +     │
  │            │    │            │    │            │    │ personal app     │
  └────────────┘    └────────────┘    └────────────┘    └──────────────────┘

  one primitive     focused product    focused product    everything in
  done deep         (business memory)  (business memory)  one API + apps


                     SUBJECT — WHAT THE MEMORY IS ABOUT

  ┌────────────────┐    ┌────────────────────┐    ┌────────────────────┐
  │ PEERS ONLY     │    │ BUSINESS ONLY      │    │ BOTH               │
  │                │    │                    │    │                    │
  │ Honcho         │    │ Orectic            │    │ Supermemory        │
  │                │    │ Penumbra           │    │                    │
  └────────────────┘    └────────────────────┘    └────────────────────┘

Four-way comparison

Supermemory wins the "broadest and cheapest to try" axis unambiguously. It doesn't necessarily win the "best at any one thing" axis — that's where each of the other three has room.

Supermemory vs Honcho — the real head-to-head

This is the interesting one. Orectic and Penumbra don't overlap much with Supermemory (they ship business-KG products, Supermemory ships context infrastructure). But Supermemory and Honcho are chasing the same customer with different theories of the product. Both ship OpenClaw plugins, both ship Claude Code plugins, both use the word "dreaming" for background reasoning. Neither references the other on their site.

Where each pulls ahead

Evidence signals

Supermemory has the strongest overall evidence stack of any product on this hub — larger distribution than Honcho, real customer logos, more compliance certifications, and they built the eval platform others benchmark against.

Distribution

Named customers

Neither Orectic, Penumbra, nor Honcho displays named enterprise customers on their site. Supermemory's positioning is anchored on real production usage.

Compliance

Benchmarks

SOTA on LongMemEval, LoCoMo, and ConvoMem per their research page. They also built MemoryBench — an open eval platform for memory systems. That's a moat: whoever runs the eval platform partly defines what "state of the art" means in the category.

Integrations for what you already run

Supermemory ships six agent plugins. Three of them (OpenClaw, Claude Code, Chrome) directly overlap with tools Organized AI uses. The OpenClaw and Claude Code plugins mean you can bench Supermemory head-to-head with Honcho on identical infrastructure.

Install (verbatim from supermemory.ai)

# One-line setup (personal / dev)
npx supermemory setup

# Personal app (macOS)
https://app.supermemory.ai

# MCP server for any MCP-speaking agent
https://supermemory.ai/mcp

None of these require touching your existing Orectic/Penumbra/Honcho evaluations. Supermemory can run alongside as a parallel bet, and the OpenClaw and Claude Code plugins are the cheapest ways to A/B against Honcho on the same workload.

Transparent, layered pricing

Supermemory has the most transparent pricing of the four products, and the only free tier. Monthly plan + consumption-based billing on top, with token-level deduplication so re-uploads don't re-bill.

Consumption model — SM tokens

On top of the monthly plan, usage is metered by SM tokens (their unit of ingested/embedded content). The trick: deduplication at the token level. Re-uploading a doc, syncing a connector, or looping the same context does not re-bill.

Startup program: qualifying early-stage startups get the Scale plan free for 3 months (~$600/mo usage) plus onboarding. Academic research teams too.

When to reach for Supermemory

The decision framework, updated for the 4-way.

Stack & conventions

Same single-file HTML pattern as the other five sibling sites. Dark terminal theme, monospace topbar, sticky sidebar. ASCII diagrams use highlight spans — this page adds blue (.hb) as an accent for Supermemory-specific badges (customer logos) to visually distinguish them from Orectic (amber), Penumbra (teal), and Honcho (purple) in shared diagrams.

Deploy & run

CLOUDFLARE_ACCOUNT_ID=691fe25d377abac03627d6a88d3eeac9 \
  wrangler pages project create orectic-penumbra-supermemory \
  --production-branch main 2>/dev/null || true

cd docs/supermemory
CLOUDFLARE_ACCOUNT_ID=691fe25d377abac03627d6a88d3eeac9 \
  wrangler pages deploy . \
  --project-name orectic-penumbra-supermemory \
  --branch main \
  --commit-dirty=true