Neotoma

Neotoma is a typed, versioned, user-controlled memory substrate that AI agents read and write through MCP. It gives cross-tool agents shared, verifiable state — not another notes app, not provider-locked chat memory.

Repo and docs are open at github.com/markmhendrickson/neotoma. Product site is neotoma.io.

• Agents are moving from chat into long-running, multi-tool work. The missing primitive isn't another retrieval-style memory — it's a substrate with provenance and reproducible state.
• Provider-native memories are account-scoped and opaque. Users running agents across models and tools end up acting as the human sync layer between them.
• MCP makes a tool-agnostic memory layer practically deployable for the first time. Open schemas plus content-addressed observations make it auditable.

• Schema-first extraction with hash-based IDs and reproducible state
• Versioned observations with field-level provenance and immutability guarantees
• Cross-tool access through MCP — works with Claude, ChatGPT, Cursor, and custom agents in one substrate
• User-controlled by default: local-first storage, typed export, user-held keys
• Composable typed primitives that span domains (work, finance, health, relationships) rather than per-vertical silos

Working product in active developer release. Used daily by the founder across a full agentic stack — code, writing, calendar, finances, and infrastructure operations.

The problem is acute enough that developers keep building their own homegrown agent-memory systems — flat-file state, JSON heartbeats, custom MCP servers, personal Postgres schemas — and surfacing them publicly. Field conversations with infrastructure engineers and operators surface the same gaps in each: no versioning, no provenance, no cross-tool access. The framings they converge on independently — “state integrity, not retrieval quality,” “CI/CD for agent state” — match the design Neotoma already ships.

Public documentation, open repo, deterministic guarantees written down. A small set of early design-partner conversations and integrations is under way with operators running personal and professional agent infrastructure.

Going to depth on the personal agentic OS builder/operator before expanding into team and pipeline contexts.

Default mode is capital-efficient and self-funded. Revenue objective for the next twelve months is in the low-six-figure range, oriented around mid-market design partners rather than volume subscriptions.

Open to small, time-boxed rounds when a profitable milestone is in sight and terms align with founder-friendly defaults: non-participating 1x, no vetoes on hiring or pricing or product, founder vesting unaltered, 50%+ diluted control preserved until durable profitability.

• Investors comfortable with a revenue-first, capital-efficient pace and a founder who treats product depth and market validation as the primary risks to retire
• Funds whose theses cover agent infrastructure, deterministic state, AI-native developer tools, or user-sovereign data primitives
• Partners who value working artifacts and open documentation over narrative-led decks

If the above resonates and there's plausible alignment, I'm happy to set up a working walkthrough — live system, current architecture, and where the business model is pointing.

Email me and please include:

• What about the above resonated and where you think there's alignment
• Your thesis area and recent investments closest to this space
• Stage and check size you typically lead or follow
• What a productive first conversation would look like