Introducing Open Context: Your AI Memory, Portable and Private

Why does AI feel so forgetful?

According to the Stack Overflow Developer Survey 2025 (49,000+ respondents across 177 countries), 84% of developers now use or plan to use AI tools, with 51% using them daily — yet trust in AI accuracy has fallen sharply, from 40% in prior years to just 33% today, with 46% actively distrusting AI outputs and only 3% expressing high confidence in AI results. 66% cite "AI solutions that are almost right, but not quite" as their top frustration. The models aren't getting worse. The problem is context — and the lack of persistent memory across sessions and tools.

Every conversation starts from zero. Claude doesn't know how you prefer to receive feedback. ChatGPT doesn't remember that you're building a Rust monorepo. Gemini has never heard of the team conventions you spent months explaining to another assistant. When you switch tools — or even start a new session — all of that accumulated understanding disappears.

This isn't a model quality problem. It's a memory architecture problem.

"Context engineering is the delicate art and science of filling the context window with just the right information for the next step." — Andrej Karpathy, via Simon Willison, 2025. The insight is simple but underappreciated: what the model knows at inference time matters as much as how the model was trained. Structured, portable context is the missing layer between AI tools and truly productive workflows.

What is Open Context?

Open Context is an open-source tool that solves the AI memory problem at the infrastructure level. It imports your full conversation history from ChatGPT, runs local AI analysis using Ollama, and produces a set of structured, portable context files that any AI assistant can consume immediately. The entire pipeline runs on your hardware — no external API calls, no cloud uploads, no subscriptions.

Open Context is a free, MIT-licensed CLI, dashboard, and MCP server that imports ChatGPT conversation history, analyzes it locally with Ollama, and generates three portable memory files — preferences.md, memory.md, and user-profile.md — compatible with Claude, ChatGPT, and Gemini. No API keys or cloud accounts required.

How does it work?

The pipeline has three stages. Getting started takes one command.

How does MCP give Claude persistent, local memory?

The Model Context Protocol has become the de facto standard for connecting AI agents to external tools — over 97 million monthly SDK downloads, 10,000+ active public servers as of December 2025 (with over 22,000 catalogued across all directories by May 2026), and native support across Claude, ChatGPT, Google Gemini, Microsoft Copilot, Cursor, VS Code, Windsurf, and Codex CLI (Digital Applied, April 2026). A 2026 Stacklok survey found 41% of software organizations already have MCP servers in limited or broad production. On December 9, 2025, Anthropic co-founded the Agentic AI Foundation (AAIF) under the Linux Foundation alongside Block and OpenAI, with Google, Microsoft, AWS, and Cloudflare as supporting members — donating MCP as a founding project. AAIF grew to nearly 150 member organizations in its first three months, making it one of the fastest-growing foundations in Linux Foundation history.

Open Context's built-in MCP server uses this standard to give Claude something it doesn't have natively: long-term memory that persists across sessions, is stored locally, and can be searched, tagged, and updated at any time. Unlike custom API integrations, MCP tools are exposed to the model directly — which means Claude can save a note, recall a preference, or search your memory store in the middle of a conversation without any extra tooling on your end.

Why does 100% local processing matter for your data?

"Privacy-first" is a phrase every AI product uses. Open Context makes it structural rather than a policy. There are no external API calls in the analysis pipeline. No telemetry. No account required. Ollama runs entirely on your hardware, which means your conversations, your inferred preferences, and your memory files never touch a remote server.

The data store lives at ~/.opencontext/ — a plain JSON directory you can inspect, back up, or delete at any time. The dashboard includes a privacy toggle that blurs all personal data in the UI, useful for screen sharing or public demos.

This matters for a specific audience: developers who work with proprietary code, healthcare professionals, lawyers, and anyone whose conversations contain information they wouldn't want leaving their machine. The local-first design isn't a limitation — it's the point.

Open Context's local processing model means no telemetry, no remote inference, and no account required. Data lives in ~/.opencontext/ — a plain JSON directory you own entirely. According to the Stack Overflow Developer Survey 2025, trust in AI accuracy has fallen to just 33% — down from 40% the year prior — with 46% of developers now actively distrusting AI outputs and only 3% expressing high confidence. Open Context is structurally designed for that skeptical majority: your data never leaves your machine, and you never have to trust a vendor's privacy policy. On-device AI is also increasingly the compliance-friendly path: the EU AI Act becomes fully applicable August 2, 2026, and GDPR cumulative fines have reached €5.88 billion — making local-first architecture a structurally safer choice.

Where does Open Context fit in the AI ecosystem?

The AI assistant market has fragmented in the best possible way. Different tools are genuinely better at different tasks — Claude for reasoning and long-form writing, ChatGPT for broad general use, Gemini for Google Workspace integration. Power users don't pick one and stop there. They use all of them, and they're constantly rebuilding context from scratch.

Open Context is for that user. It's not a replacement for any AI assistant. It's the infrastructure layer between them — the place where your preferences, history, and working context live, independent of any single platform.

Open Context is the portable memory layer between AI assistants — not a replacement for Claude, ChatGPT, or Gemini, but the infrastructure that carries your preferences and history across all of them. With MCP now governed by the AAIF under the Linux Foundation and supported natively by every major AI platform, the standard for AI interoperability is settled. Open Context is built for that world — with 10,000+ active MCP servers deployed and 41% of software organizations already running MCP in production, the infrastructure your memory layer connects to is already here (Digital Applied, April 2026).

The project is MIT-licensed and built on a stack that experienced developers will find familiar: Node.js, TypeScript, Express, React + Vite, and Commander.js for the CLI. Contributions are welcome. Gemini import support is on the roadmap. If you're looking to go further — running entire organizational functions with AI or validating agent reliability before shipping — see our posts on Open Org and ARA.

Frequently Asked Questions

Getting started with Open Context

Open Context is live on GitHub and available now. The quickest path is Docker — clone the repo, run docker-compose up, and the dashboard is running at localhost:3000 within minutes.

If you want persistent memory for Claude specifically, start with open-context mcp start and follow the MCP configuration guide in the README. The context migration pipeline — import, analyze, export — takes about ten minutes to run end-to-end on a modern machine.