Privacy Policy

1. Information We Collect

We collect information that you provide directly to us, including but not limited to:

• Account information (name, email address)
• Chat conversations and agent runs in the web application
• Files and documents you upload (for example, to DataHub)
• Billing information and usage metering (processed by Stripe; we never store card numbers)
• Usage data and analytics, collected via PostHog with cookies
• Technical information about your device and system

2. Your Code and the CLI

The Memcode CLI runs on your machine. Your source code and its memory stay local; we do not upload or store your repository. When the agent needs a model, the context relevant to that request is sent through the Memcode gateway to the model provider. The gateway is stateless: it holds prompts only in transit and does not persist your code or conversations. The web application stores only what you put in it.

3. How We Use Your Information

We use the information we collect to:

• Provide and maintain our services
• Route your requests to AI model providers and return responses
• Meter usage and process billing
• Analyze usage patterns and optimize performance
• Communicate with you about updates and changes

We do not use your code, conversations, files, or any other customer data to train AI models, ours or anyone else's. This applies to all Memcode products except Halo, which may use eligible, privacy-reduced interaction and feedback records to improve its language understanding, reliability and safety, as described in section 9. Halo never uses your files, their contents, command output or credentials.

4. Sign-In Data

You sign in to Memcode with Google or GitHub via OAuth 2.0. We receive only your basic profile information (name, email address, avatar) to create and identify your account. We do not request access to your Gmail, Google Drive, calendar, repositories, or any other data on those platforms. You can revoke Memcode's access at any time from your Google Account permissions or GitHub application settings. Any use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.

5. Information Sharing

We do not sell or rent your personal information or your data to third parties. We share information only in the following circumstances:

• With your consent
• With AI model providers (OpenAI, Anthropic, Google, xAI, and Fireworks) to process the requests you make — these providers process data under enterprise API terms that prohibit retaining your data for model training
• With the infrastructure providers that run our service (Vercel, Supabase, Google Cloud, Stripe, PostHog), each acting as a processor on our behalf
• To comply with legal obligations
• To protect our rights and safety
• In connection with a business transfer or acquisition

6. Data Security

We implement appropriate technical and organizational measures to protect your personal information against unauthorized access, alteration, disclosure, or destruction. These measures include encryption of data in transit and at rest, authentication via OAuth 2.0, and access controls on all stored data. If you bring your own provider API keys, they are encrypted in a dedicated secrets vault, are never written to the application database or logs, and can be read only by the service that routes your requests. Customer data is hosted in the United States. See our Security page for details.

7. Data Retention

We retain your account data and chat history for as long as your account is active. When you delete a chat, file, or your account, the associated data is permanently removed from our systems. Billing records are retained as required by law.

8. Your Rights

You have the right to:

• Access your personal information
• Correct inaccurate data
• Request deletion of your data
• Object to data processing
• Request data portability
• Revoke sign-in access at any time

To exercise any of these rights, contact us at the address below.

9. Halo

Halo (in Preview) lets you operate your computer in plain language. It runs commands on your computer and uses the hosted Halo service, operated by Memcode, to interpret what you ask.

What Halo sends to the Halo service

• The text of your request, and the earlier requests of the same conversation with how each turned out
• Context needed to interpret it: your operating system, its version and processor type, your shell, the current folder's path, your home folder path and user name, the time, time zone and locale, which common tools are installed, and whether the folder holds common project files
• Limited metadata about results you are working with (their kind and number), never their contents
• Your account identity, to authenticate you and apply usage limits

What stays on your computer

Commands run on your computer, under your confirmation where Halo asks for it. File contents, folder listings, command output, results, environment variable values, credentials and documents are never sent to the Halo service.

Operational and security data

To run, secure and troubleshoot the service, we keep operational records: request counts, the kind of response, the operations proposed, outcomes, timings and error categories, with your account identified by a pseudonym. These records do not contain the text of your requests. We use them to provide the service, detect abuse and keep it reliable, not to train models. Our infrastructure also keeps standard security and access logs (such as IP addresses at our network edge) for a short period.

How Halo improves

Halo may use eligible, privacy-reduced interaction and feedback records to improve its language understanding, reliability and safety. The most useful records are moments where Halo was unsure or wrong: a clarification you chose, a correction you gave, a preview you rejected, and requests Halo could not do yet. Raw terminal sessions are not collected, and nothing goes directly from your use into a model.

Improvement records are created only from privacy-reduced data, kept separately from operational records, and governed separately:

• Before anything leaves your computer for this purpose, Halo replaces the specific values in your request and in Halo's interpretation (paths, file and folder names, names, addresses, text) with typed placeholders, and removes your home folder, user name, email addresses and anything that looks like a credential. The service checks again and rejects records where identifying text remains.
• Records do not contain your account or user identifier, email address, IP address, hostname or device identifiers, and keep the date but not the time.
• They carry a pseudonym that changes periodically, used only to limit how much any one person can influence Halo and removed before training. Records are therefore pseudonymized, not anonymous.
• Records are reviewed and filtered before use. Halo learns from patterns across many people, not from one person's habits.

Never used to improve Halo: your files and their contents, command output, folder listings, environment variable values, credentials and secrets, and documents; data sent before this version of this policy; and, by default, data from business accounts.

Your choices

• Improvement use is on by default. Turn it off at any time with /improvement off in Halo; after that, no improvement records are created from your use.
• Diagnostics are a separate setting (/telemetry) and never include the text of your requests.
• Business accounts are excluded from improvement use by default.
• You can ask us to delete your Halo data, including improvement records that can still be linked to you, using the contact below.

Retention. Each kind of data has its own retention period, and none is kept indefinitely. Operational and security records are kept for [TO SET]. Halo does not store raw request text for improvement; privacy-reduced improvement records are kept for [TO SET]. Training datasets built from them are kept only while the models trained on them are in use.

Providers. The Halo service and its records are hosted on Google Cloud in the United States, acting as our processor.

10. Contact Us

If you have questions about this Privacy Policy, please contact us at tim@memcode.ai