NinaiInstall Ninai
Menu

Local-first memory for the AI tools you already use

Your AI should remember the work.Not your whole life.

Ninai saves the decisions, commitments, and project context that should survive a session—on your device. Each assistant recalls only what you allow.

Install Ninai View the source
  • One local install
  • No duplicate connectors
  • No account

The context aperture Try the permission switch

Local vault37 durable memories
On this device
Claude Code can recall
01
Finish the permission dashboard.linear://NIN-42
02
Keep Linear connected where it is.decision://capture-architecture
2of 37 memories released
PermissionProject memory
2 facts visible

What Ninai changes

A session ends. The useful part shouldn’t.

Today, an AI can read your tools and still forget the outcome tomorrow. Ninai keeps the durable result, not the entire transcript.

Session 01Claude Code + Linear

“Review NIN-42 and tell me what matters.”

Linear returnedPermission dashboard required before launchowner · deadline · status · source
Session 02Fresh context

“What must I finish before launch?”

01
Finish the permission dashboard.linear://NIN-42 · confidence 0.97

1 useful fact · source included

One install

Keep your tools where they are.

Linear remains connected to Claude. GitHub remains connected to Claude. Ninai listens to completed host events, so you do not re-enter credentials or rebuild the setup you already trust.

See the four-minute setup →
Already connected
LI Linear MCP
GH GitHub MCP
•• Your tools
existing calls
AI hostClaude CodePostToolUse event
local hook
Your machineNinaicapture · policy · recall

Who it is for

Different work. The same missing continuity.

Ninai starts with people who rely on AI for real decisions and cannot afford to rebuild the story in every new session.

01Builder

“What did we decide about the auth flow?”

Stop repeating technical decisions.

Architecture choices, project state, and completed tool outcomes return in a fresh coding session.

Existing GitHub and Linear MCP connections stay untouched.
02Operator

“Who is waiting on me this week?”

Keep commitments in view.

Deadlines, owners, follow-ups, and decisions survive the context switches that normally erase them.

Every returned outcome keeps its original source.
03Steward

“What did Claude actually receive?”

See and control the boundary.

Inspect the exact packet, grant only useful scopes, and revoke future access without deleting the memory.

Permission is evaluated before retrieval begins.

PACT retrieval

Small by design.

More context is not automatically better context. Ninai first narrows memory by permission, then returns the smallest evidence packet that can still help.

  1. 01
    Permit

    Resolve the client’s explicit scopes.

  2. 02
    Select

    Rank only permitted, current evidence.

  3. 03
    Return

    Fit useful facts and sources into the budget.

Read the research direction →
Context packet263 tokens
PurposePrepare today’s Ninai work
A
Finish the permission dashboard.linear://NIN-42
B
Keep existing MCP connections in place.decision://capture-architecture
2facts returned35memories stayed private

The honest boundary

The vault stays. The packet travels.

Your complete memory remains on your machine. If Claude or another cloud AI uses a selected packet, that packet leaves the device and the provider’s policy applies.

Inspect the privacy architecture →
Your machineComplete local vault37 memories · permissions · sources · logs
permission2 facts
Selected AICompact packetProvider policy applies after release

Local MVP

Try the boundary yourself.

Install the Python engine, grant Claude Code one project scope, remember a fact, recall it, then revoke the scope. No account required.

Open the install guide 
cd engine
python3 -m venv .venv
source .venv/bin/activate
pip install .

ninai permission grant claude-code project
claude mcp add ninai -- ninai-mcp

Questions

Clear before clever.

Do I reconnect Linear, GitHub, or other MCP tools?

No. The Claude Code integration observes completed tool activity through a PostToolUse hook. Your existing MCP configuration stays where it is.

Does Ninai send my complete memory vault to an AI?

No. Ninai filters by the requesting client’s granted scopes first, then releases a compact packet containing only selected facts and their sources.

Where is memory stored?

The current MVP stores memory, provenance, permission grants, and disclosure logs in a local SQLite database on your machine.

Is Ninai production encrypted?

Not yet. The MVP is honest about this boundary: it uses local SQLite and does not claim SQLCipher, an independent audit, or production security certification.

Remember the work.Keep the rest yours.

Install Ninai ↗Talk to the founder