SutraStart with Sutra

Private context infrastructure

Give every AI the context you choose.

Sutra is a shared, private context layer for your AI tools. Store durable decisions and preferences once, then carry them between compatible agents without starting over.

Create your private workspaceSee context move

One mind. Every AI. Always under your control.

Sutra
Connected

Memory index

What your agents know

Review the durable context available across connected clients.

All systems ready

The public launch should lead with shared private context, not generic AI productivity.

#positioning#launch

Product direction

via Claude Desktop

Keep implementation updates direct: outcome first, supporting detail second.

#communication

Working preferences

via ChatGPT

Context remains inspectable and under your control.

Tested across 22 agents

View all
  • Claude.ai
  • ChatGPT
  • Lovable
  • Replit
  • Bolt.new
  • v0.dev
  • OpenAI Codex
  • Grok
  • Devin
  • Cursor Web
  • Notion
  • Claude Code
  • OpenCode
  • GitHub Copilot
  • Gemini CLI
  • Cursor Desktop
  • Windsurf
  • Cline
  • Aider
  • Continue.dev
  • Zed AI
  • Copilot Chat for VS Code

Why Sutra

Your context should outlast the chat window.

Stop retelling your tools

Carry stable preferences, project constraints, and decisions into the next conversation, even when the client changes.

See exactly what persists

Search memories, inspect their source, and organize context into clusters instead of trusting an invisible black box.

Connect tools, not silos

Use the same private context layer through supported MCP and API connections rather than maintaining separate AI profiles.

Real workflow

The thread continues across tools.

A connected agent writes the durable part. Sutra keeps it private and inspectable. Another connected agent can retrieve it when relevant.

  1. Claude

    Connected client

    Remember that the launch leads with private shared context, and team billing stays out of scope.

  2. Sutra

    Private context layer

    Stores the durable decision in Product direction, with its source and importance visible in your workspace.

  3. ChatGPT

    Another connected client

    Picks up the same launch constraints without making you rebuild the context from scratch.

Sutra shares durable memories that connected agents write and read. It does not copy entire conversations between clients.

Portable skills

Upload once. Enable with intent.

Skills package reusable instructions and supporting resources for an agent. Memories are different: they preserve the decisions, facts, and preferences that make the work yours.

Nothing becomes available by accident.

Every uploaded or edited version starts Disabled and remains immutable.

Research synthesis

Reviewed immutable version / v3

Enabled

Skill bundle

  • SKILL.mdInstructions
  • resources/
  • rubric.md
  • template.md

One root SKILL.md, with optional private resources in the same reviewed bundle.

  1. Upload once

    Humans upload in the dashboard. Connected agents can submit drafts, but agent-submitted versions stay Disabled.

  2. Review the immutable version

    Inspect SKILL.md and every resource. Uploads and edits create a new version instead of changing an existing one.

  3. Explicitly enable it

    Choose the reviewed version agents may retrieve. Disable it again without deleting the skill or its history.

One enabled version, available across connected clients.

Claude, Copilot, OpenCode, and other MCP clients can retrieve the same approved SKILL.md and resources through Sutra.

Install the CLI

From zero to shared context in one command.

Connect Claude Code or OpenCode without hand-editing configuration. Sutra merges safely and installs the instructions each agent needs for the next session.

Open connection setup
Terminal
Node 18+
Install globally
npm install -g @shivnathtathe/sutra-cli
Connect Claude Code
sutra connect claude-code
Connect OpenCode
sutra connect opencode
connected  Claude Code
connected  OpenCode

Security and control

Private memory with an off switch.

Context is useful only when you can govern it. Sutra gives every connection an identity and gives you a workspace to review what remains.

Scoped connections

Issue separate API keys for each agent or environment so access is not shared indiscriminately.

Independent revocation

Turn off one key without disrupting the rest of your connected workspace.

Inspectable by design

Review memory content, source agent, cluster, tags, and activity from your own dashboard.

Compatible clients

Meet your agents where they work.

Connect Sutra through MCP or its API. Client support depends on the connection methods each AI product exposes.

Claude Desktop and Claude Code
Cursor and MCP-compatible agents
ChatGPT and custom integrations

One private context layer

Start the next conversation with what already matters.

Create your private workspace