memside

# Memside

Memside is an AI continuity and memory hub for work across ChatGPT, Claude, Grok, IDEs, coding agents, and other AI tools that support OAuth, MCP, or API-key connections. It provides a private place for reusable context, decisions, Operating Rules, project notes, checkpoints, and Library Templates that can move between compatible AI tools.

Memside keeps AI context available beyond one chat window. A connected AI tool can retrieve relevant memories, continue a project from a checkpoint, follow saved working preferences, create approved durable context, and work with permission-controlled Library Templates. This repository is the public home for Memside integration documentation, examples, SDKs, and support information. It is not a mirror of the private production application source.

Product site: [https://www.memside.com](https://www.memside.com)

Documentation: [https://docs.memside.com](https://docs.memside.com)

## Continuity With Less Repeated Prompting

Memside is built to give AI assistants the right context at the right time:
current task state, saved preferences, project notes, checkpoints, and reusable
rules. In internal testing and early pilot usage, this helped AI tools stay on
track without requiring users to rebuild the same detailed prompt in every new
chat or IDE session.

High-level test results:

| Context approach | Continuity score |
| --- | ---: |
| No Context | ~45% |
| Long Context Dump | ~84-85% |
| Memside Context | ~96-99% |
| Very Detailed Manual Prompt | ~97-99% |

**Token savings:** in separate context-reuse tests, Memside reduced repeated
full-context transfer by **83-91%** on representative light, medium, and heavy
project datasets.

A very detailed manual prompt means a long prompt where the user includes every
important detail by hand: project state, preferences, decisions, references, and
next steps. With Memside, the workflow is simpler: ask the AI to save the
checkpoint, references, and useful context, then move to a new chat and ask it
to continue from that checkpoint using Memside.

Results are based on internal tests and early pilot usage patterns. Performance
varies by model, provider, prompt style, and workflow. See
[Continuity and context results](docs/continuity-results.md) for details.

## Connect Memside

Memside supports hosted MCP access for AI tools that can connect to remote MCP servers. MCP gives compatible AI clients a standard way to use Memside as a continuity layer instead of locking memory into one assistant.

```text
https://api.memside.com/mcp/
```

For AI tools that support OAuth for MCP, setup continues through Memside sign-in. AI tools that support API-key authentication can use a Memside API key as a bearer token.

Start here:

- [Official documentation](https://docs.memside.com/)
- [Developer quickstart](docs/developer-quickstart.md)
- [JavaScript SDK](packages/memside/README.md)
- [Python SDK](packages/python/README.md)
- [Capabilities](docs/capabilities.md)
- [Continuity and context results](docs/continuity-results.md)
- [Token savings and context reuse](docs/token-savings.md)
- [Memory types and sensitivity](docs/memory-types.md)
- [Connect AI clients](docs/clients/README.md)
- [Client support matrix](docs/client-support.md)
- [Install in VS Code](docs/clients/vscode.md)
- [Use Memside with API keys](docs/api-keys.md)
- [Public API reference](docs/public-api.md)
- [Memside Library and Templates](docs/template-library.md)
- [Curated OpenAPI document](openapi.json)
- [MCP Registry metadata](docs/mcp-registry.md)
- [Directory metadata](docs/mcp-registry.md#third-party-directory-metadata)
- [Troubleshooting](docs/troubleshooting.md)

## Developer Quickstart

Select the path that matches the client or integration.

| Goal | Use | Start here |
| --- | --- | --- |
| Connect ChatGPT, Grok, Claude, or another hosted client | OAuth MCP | [Client guides](docs/clients/README.md) |
| Connect IDEs and coding agents such as VS Code, Copilot, Cursor, Claude Code, Codex, Antigravity, opencode, or a local MCP client | Bearer-header MCP | [VS Code guide](docs/clients/vscode.md) |
| Call Memside from scripts or automation | API key | [API keys](docs/api-keys.md) |

For a direct API check, create a Memside API key and run:

```bash
curl https://api.memside.com/context/startup \
  -H "Authorization: Bearer mem_sk_example_key"
```

For a runnable JavaScript example:

```bash
cd examples/javascript
npm run startup
```

Set `MEMSIDE_API_KEY` before running examples. See [Developer quickstart](docs/developer-quickstart.md) for the full setup.

## JavaScript SDK

Memside publishes a lightweight JavaScript SDK for public API-key workflows:

```bash
npm install memside
```

Basic usage:

```js
import { MemsideClient } from "memside";

const memside = new MemsideClient({
  apiKey: process.env.MEMSIDE_API_KEY
});

const startup = await memside.context.startup();
console.log(startup);
```

The JavaScript SDK wraps public Memside API routes for startup context, resume
context, workspace profile, Memory operations, Subject organization, and
Library Template workflows. See
[packages/memside](packages/memside/README.md) for package details.

## Python SDK

Memside also publishes a lightweight Python SDK for public API-key workflows:

```bash
pip install memside
```

Basic usage:

```python
from memside import MemsideClient

client = MemsideClient(api_key="mem_sk_example_key")

startup = client.context_startup()
print(startup)
```

The Python SDK wraps the same public Memside API areas as the JavaScript SDK. See [packages/python](packages/python/README.md) for package details.

## MCP Server Setup

Use this hosted MCP server URL in clients that support remote MCP:

```text
https://api.memside.com/mcp/
```

For ChatGPT, Grok, Claude, and other clients that offer OAuth during MCP setup, choose OAuth and complete Memside sign-in.

For clients that use bearer headers, create a Memside API key and pass it as:

```text
Authorization: Bearer mem_sk_example_key
```

VS Code and GitHub Copilot can use a workspace `.vscode/mcp.json` file:

```json
{
  "servers": {
    "memside": {
      "type": "http",
      "url": "https://api.memside.com/mcp/",
      "headers": {
        "Authorization": "Bearer mem_sk_example_key"
      }
    }
  }
}
```

Keep real API keys out of shared files. Use local user-level configuration or secret storage when the config is shared with a team.

Detailed setup guides are available for [VS Code](docs/clients/vscode.md), [GitHub Copilot](docs/clients/github-copilot.md), [ChatGPT](docs/clients/chatgpt.md), [Grok](docs/clients/grok.md), [Claude Code](docs/clients/claude-code.md), [Cursor](docs/clients/cursor.md), [Codex](docs/clients/codex.md), [Antigravity](docs/clients/antigravity.md), and [opencode](docs/clients/opencode.md).

## MCP Registry

This repository includes `server.json` for the official MCP Registry. It describes Memside as a hosted remote MCP server using Streamable HTTP at:

```text
https://api.memside.com/mcp/
```

The registry entry is published under:

```text
io.github.memside/memside
```

See [MCP Registry metadata](docs/mcp-registry.md) for registry and endpoint checks.

This repository also includes lightweight third-party directory metadata for Glama, Open Plugins, and Cursor plugin discovery. These files point to the same hosted MCP server URL and do not contain credentials.

## Supported Workflows

With Memside connected, an AI tool can search saved memories, fetch specific context, create approved memories, update checkpoints, and build compact startup or resume packets. The goal is to keep useful context portable without turning every AI conversation into a long manual setup step.

This supports project movement between ChatGPT, Claude, Grok, IDEs, coding agents, and other compatible AI clients. The next AI can retrieve the project purpose, prior decisions, applicable rules, and current next step.

Memside Library adds reusable Templates for profiles, Operating Rules, AI
Skills, memories, references, examples, and repeatable workflows. Public
catalog discovery is read-only. Permission-controlled creator operations can
search Templates, read owned drafts, replace drafts safely, and manage reviewed
publication workflows. See [Memside Library and Templates](docs/template-library.md).

Memside keeps the normal product boundary intact. Private user data is scoped to the signed-in user, secret memories are excluded from AI-facing MCP and API-key flows, and destructive actions require stronger confirmation.

## Key Features

Memside is designed around continuity, selective context retrieval, and user control. The core features are:

- [Startup and resume context](docs/token-savings.md) for compact handoffs.
- [Operating Rules](docs/operating-rules.md) for durable behavior guidance.
- [User AI Profile](docs/user-ai-profile.md) for personal working preferences.
- [AI Skills](docs/ai-skills.md) for reusable instruction packs.
- [Public, private, and secret memories](docs/memory-types.md) for practical data boundaries.
- [MCP and API-key access](docs/clients/README.md) for connecting multiple AI clients.
- [Memside Library and Templates](docs/template-library.md) for reusable AI setup and creator workflows.

## API Examples

The examples in this repo use public Memside API surfaces only.

- [curl examples](examples/curl/README.md)
- [JavaScript examples](examples/javascript/README.md)
- [Python examples](examples/python/README.md)

The JavaScript and Python examples use normal HTTP requests against public Memside API surfaces.

The machine-readable [OpenAPI document](openapi.json) contains only the
approved API-key contract. It excludes signed-in application, account,
administration, billing, and infrastructure operations.

## Public Repo Scope

This repo contains:

- setup guides for MCP clients
- MCP Registry metadata
- third-party MCP directory metadata
- Cursor plugin metadata
- technical product capability docs
- API-key usage examples