Your AI memory, in plain Markdown

Stop re-explaining yourself to every AI

Context OS turns your ChatGPT, Claude and Gemini history, plus your docs and notes, into one private memory in plain Markdown. Claude Code, Cursor, ChatGPT and any MCP agent know you from the first message.

Start freeOr try the instant demo, no sign-up

Free audit and a 150K-token taste build. No credit card.

  • Secrets scrubbed in your browser before upload
  • Plain Markdown you own, no lock-in
  • Any language, mixed sources welcome
Works withAny AI / MCP agentClaude CodeCursorWindsurfDeepSeekQwenKimiGrokChatGPTGeminiClaude DesktopObsidian
My Context Vault3 secrets scrubbed

Your exports, checked in your browser

  • conversations.json

    ChatGPT export · 1,284 conversations

  • claude-export.zip

    Claude export · 412 conversations

  • Raseed-pricing.pdf

    Document

Tokens2M420K79% removed for free before any AI call
Illustrative example built from the demo vault (Layla Haddad’s sample data).

The Token Wall

Pasting your history into a chat can’t work

A chat model only sees what fits in its context window. Years of conversations do not — so the model quietly forgets most of what you told it.

Paste raw exports into a chat

Your raw export2,000,000 tokens
A typical context window128K–200K tokens

Around 90% of your history never reaches the model.

  • Older facts are cut off or silently truncated
  • Outdated and current decisions contradict each other
  • You pay again every time you re-paste the same context

Distill once, route on demand

After free pre-LLM filtering~70–85% fewer tokens
What an agent loads per taskThe map + only the files it needs

Every fact is kept — agents read just the slice the task needs.

  • Noise, trivial chats and duplicates removed locally for $0
  • Newest fact wins; older versions move to a Historical Archive
  • Agents read files on demand over MCP instead of a giant prompt

Illustrative example. Your numbers depend on your exports; the free audit shows your real counts before you pay anything.

How it works

From messy exports to a vault your agents can navigate

Six engines run in order. The expensive step — the AI — only ever sees clean, deduplicated, secret-free text.

  1. 01

    Free browser audit

    Your files are opened inside your browser. You see conversation counts, raw tokens, the dominant language and a preview of your folder tree — at $0 and before checkout.

  2. 02

    Local secret scrubbing

    API keys, passwords, IBANs and card numbers become placeholders like [REDACTED_SECRET_1]. The real values go only into a local .env.example.

  3. 03

    Chronological distillation

    Sources are processed oldest to newest. When a newer message contradicts an older one, the newest truth stays and the old fact is archived with its date.

  4. 04

    Agent routing maps

    AGENTS.md, CLAUDE.md, GEMINI.md, SOUL.md, .cursorrules and folder README.md files with IF/THEN rules, in compact English that every agent reads efficiently.

  5. 05

    Self-healing structure

    Rename or move anything and every map updates instantly at zero token cost. Create a folder and the Contextual Magnet pulls in matching files — with one-click undo.

  6. 06

    Connected everywhere

    An MCP endpoint for Claude Desktop, Cursor, Windsurf and ChatGPT, a browser extension for web chats (early access), and loop-safe sync to a folder on your drive.

Hybrid privacy architecture

Private by design, and precise about how

Four layers protect your context. Here is exactly what each one does, including the one step where our servers process your sanitized text.

  1. 1

    Client-side secret scrubbing

    API keys, passwords, cards and IBANs are replaced with placeholders by code running entirely in your browser, before anything is uploaded. The real values go only into a local .env.example.

  2. 2

    Ephemeral in-memory distillation

    The sanitized, secret-free text is distilled by the AI in server memory. Staged uploads are deleted when the job ends. With your own AI key, the model call runs on your provider account.

  3. 3

    Encrypted at rest

    Vault files and Quarantine diffs are stored encrypted with a per-vault AES-256-GCM data key (envelope encryption), so agents can still use the vault over MCP.

  4. 4

    Optional zero-knowledge lock

    Add a passphrase and your browser encrypts every file with a key we never receive (PBKDF2 + AES-256-GCM). While locked, we hold only ciphertext, and AI, MCP and the extension pause until you unlock.

Tier 1 · Managed cloud

Managed, with hybrid encryption

Hosted by us with hybrid encryption: secrets are scrubbed before upload, files are encrypted at rest, and you can lock storage to zero-knowledge.

  • Secrets and keys are scrubbed in your browser before upload
  • Sanitized text is distilled in server memory; staged uploads are deleted after each job (sources waiting for more tokens are kept encrypted for up to 7 days)
  • Optional passphrase lock: zero-knowledge storage at rest (AES-256-GCM, PBKDF2)
  • Envelope encryption at rest keeps MCP-connected vaults usable

Tier 2 · Your own storage

Bring your own storage

Your vault is mirrored to storage you control.

  • Sync to a private GitHub repository
  • Or to AWS S3, Cloudflare R2 or any S3-compatible bucket
  • Storage credentials are stored encrypted
  • Add your own AI key so model calls run on your provider account

Tier 3 · Self-hosted

Self-hosted enterprise node

The whole stack inside your own infrastructure: the only option where no plaintext ever reaches our servers.

  • One Docker Compose file: app, PostgreSQL, S3 storage and MCP gateway
  • Optional Ollama for fully offline local models
  • Role-based folders for teams (owner, admin, member)
  • Expiring, folder-scoped links for freelancers

Accuracy note: the managed cloud is not zero-knowledge while your vault is being distilled or served to agents. The AI needs readable, secret-free text for that, and it is processed in server memory. Bringing your own storage changes where the vault is stored, not where distillation runs. If no plaintext may ever reach our servers, self-host the full stack (optionally with Ollama for local models).

Pricing

Start free. Pay only when the AI does the work.

The audit and a taste build are free. Paid plans share one pooled token budget across all your vaults, with a hard cap enforced before every AI call.

Free

$0

Free forever

150K-token taste build (one-time)

≈ 50 recent conversations distilled

Audit your exports for $0, then see real distilled Markdown before you pay.

Start free

1 vault with routing maps

Read-only MCP for your AI tools

Browser extension for web chats (early access) (not included)

Sync to your own GitHub repository or S3 bucket (not included)

Folder roles and expiring share links (not included)

1 seat

Pro

Most popular

$15/month

Billed monthly, cancel anytime

3M tokens / month, pooled

≈ 1,000 conversations or 330 chat questions a month

A living vault that keeps learning from every chat.

Start Pro

Up to 3 vaults

Full MCP (read + write) for Claude, Cursor and ChatGPT

Browser extension for web chats (early access)

Sync to your own GitHub repository or S3 bucket

Folder roles and expiring share links (not included)

1 seat

Team

$60/month

Billed monthly, cancel anytime

10M tokens / month, pooled (hard cap)

≈ 3,300 conversations a month for the team

Shared team vaults with roles, for up to 3 people.

Start Team

Up to 25 vaults

Full MCP (read + write) for every seat

Browser extension for web chats (early access)

Sync to your own GitHub repository or S3 bucket

Folder roles and expiring share links

3 seats included, more on request

Extra Team seats cost $20 per seat per month. Seats cannot be bought in checkout yet: contact us and we add them to your team.

Add-on

Import Pack

$29 one-time

Import years of history at once, on any plan.

  • +10M distillation tokens that never expire (≈ 3,300 conversations)
  • 1 month of Pro included: full MCP, browser extension, your own storage
  • No subscription; stacks with any plan
Get the Import Pack
14-day refund policy

Self-hosted

Enterprise & Self-Hosted

From $2,000 / year

or a custom quote for your organization

Run the full Context OS stack inside your own infrastructure. Vault data stays on your servers, and with local models (Ollama or vLLM) none of it leaves your network.

  • Docker self-hosted deployment (app, PostgreSQL, MCP gateway)
  • Offline signed license key, renewed yearly
  • Unlimited BYOK and local models (Ollama / vLLM)
  • Custom team seats and folder-level roles
  • SLA and priority support
Request Self-Hosted Licenseor email sales@contextos.dev

Filtered tokens are counted after free local scrubbing and filtering. Monthly allowances reset on your billing date and are shared by all your vaults; Import Pack tokens never expire. Calls made with your own AI key do not count toward the quota. Conversation equivalents are estimates. Prices in USD, plus applicable taxes.

Trust

Who runs Context OS, and how your data is protected

The facts behind the service: the company that operates it, the security measures in place, and the documents that bind us.

Security measures

  • Secrets, keys and card numbers are scrubbed in your browser before upload
  • Every vault is encrypted at rest with its own AES-256-GCM key
  • An optional passphrase lock encrypts each file in your browser with a key we never receive
  • API tokens and share links are stored only as salted hashes

Your data, your rules

  • Export your vault as Markdown files or a ZIP at any time
  • Delete your account and its vaults yourself in Settings
  • Data is processed mainly in the United States, and every subprocessor is listed
  • Team plans are covered by a Data Processing Addendum
Privacy policyData Processing AddendumTermsRefund policy

Questions

Frequently asked questions

Straight answers about privacy, tokens, and how your agents use the vault.

Get started

Give every agent your full context

Run the free audit on your exports, see your Token Wall and your future vault — then decide.

Create your free accountTalk to us

Stay in the loop

Product notes and launch updates — no noise.