anuma vs innernet

anuma is the closest thing to us in spirit, and the furthest in architecture.

"One memory. Every model. Always private." We'd sign that sentence. But anuma delivers it by being the place you chat — a private, multi-model client with an encrypted on-device memory profile. The memory is one memory because everything happens inside anuma.

innernet delivers it by not being a place you chat at all. The memory lives outside every tool and the tools you already use read it. That's the entire difference, and it decides what happens the day you want to use something anuma doesn't contain.

A private AI client with memory underneath. Access to ChatGPT, Claude, Gemini, DeepSeek, Kimi and open models through one interface, with mid-chat model switching via Council Mode. A unified memory profile stored encrypted on-device that syncs across your devices and survives model switches. Image, audio and video generation, a workflow builder for mini-apps, task agents, zero-retention open models. Web, iOS, Android and SMS/iMessage. Free and Pro tiers.

The privacy posture is local-first and genuinely strong: conversations stored on your device rather than their servers, and you can add, edit or delete anything in the profile at any time.

A memory layer under the tools you already use. npx innernet signs you in, detects the AI tools installed on your machine and connects each to the same memory — Claude Code, Cursor, Codex, Windsurf, Gemini CLI, Claude Desktop, VS Code, Cline, Zed, Continue — plus an MCP endpoint for anything else. netti files what accumulates into Context Maps with version history you can branch and roll back.

We don't ship a chat interface, we don't resell models, and we don't want to be where you work. We want to be what your work is read against.

anuma's memory is excellent everywhere anuma is. The question is what happens in Claude Code at 2am, in Cursor mid-refactor, in a terminal agent, in a tool that ships next year. anuma's honest answer is "bring that work into anuma." Ours is "we'll be there already." A memory layer's value is decided almost entirely by where it isn't.

A memory profile that persists across chats is a big improvement on starting cold. But a profile built from what you typed into a chat client is a record of your conversations. innernet reads what accumulates across actual work — sessions, decisions, the reasoning behind them — and netti writes conclusions no individual conversation contained, then updates them when the work contradicts them. Persistence and understanding are different achievements.

anuma's is protection from outsiders: encrypted, on-device, local-first, zero retention. Ours is consent toward the model: every fact carries how far it may travel — private, self, trusted, work, public — enforced when a tool reads, so private and self-only facts never leave your Self Map even with an AI connected. Both are real. anuma's is stronger against a breach. Ours is stronger against oversharing with the model itself.

• ·On-device storage. For anyone whose bar is "it never touches a server," that's the end of the conversation and we can't match it.
• ·It's a finished consumer product. Council Mode, generation, mini-apps, agents, SMS. Broad and polished.
• ·Model breadth in one place. If you want to compare answers across five models without five subscriptions, that's a real convenience.
• ·Mobile and messaging. iOS, Android and iMessage reach places we don't.

• ·It works where you already work. The memory travels to your tools instead of asking your work to travel to it.
• ·We own no model, and never will. Neutrality is easy to claim and hard to hold when your revenue depends on routing tokens. Ours doesn't.
• ·Something is reading, not just storing. netti writes and rewrites conclusions as the work changes them.
• ·Change is first-class. Branch a direction, park it, roll it back. "Edit or delete anything" is control; version history is memory.

Use anuma if you want one private, beautiful app to do your AI work in, and you're happy for that app to be where the work happens.

Use innernet if your work is already spread across editors, terminals and clients you're not going to give up, and you want all of them to know the same things about you.

Both is fine. anuma speaks MCP-adjacent surfaces; innernet is a URL.

"Anuma keeps my memory on my device. Yours is on a server. Why is that acceptable?"

Because the alternative to a server is a memory that can't read across your machines, can't be reached by a tool on someone else's laptop, and can't have anything running on it that thinks about you while you're asleep. That's the trade and we won't pretend otherwise. What we do instead: row-level security enforced in Postgres rather than in the app, tokens stored hashed, OAuth 2.1 so no password passes through us, optional column-level encryption behind a passphrase only you hold, a 90-day audit log visible only to you, and export-everything / delete-everything on demand — deletion immediate, backups purged within 30 days. Never sold. If your threat model still says on-device, anuma is the honest recommendation.

Accurate as of 3 September 2026, based on anuma's public site. If we've described their architecture or privacy model incorrectly, tell us and we'll correct it.