screenpipe

screenpipe continuously captures desktop context and turns it into searchable history for humans and AI agents. The current product combines accessibility-tree text with screen frames and OCR fallback, records microphone and system audio, stores local history in SQLite, and exposes that data through a local REST API, MCP server and SDK.¹²

The agent layer is broader than passive recall. Scheduled “pipes” can query captured history and run approved follow-up workflows, while MCP lets compatible assistants such as Claude, Codex and Cursor retrieve prior context instead of relying only on the current prompt.³² That makes screenpipe unusually programmable for a desktop-memory product, although the practical value depends on capture quality, permissions, storage settings and the agent workflows built on top of it.

The current web pricing surface includes a Free tier for one-device capture and limited search, Basic at $25/month or $250/year, and Business at $50 per seat per month or $500 per seat per year. Enterprise is contact-led on the live enterprise page.⁴ Annual pages also display effective monthly equivalents, so those lower figures should not be mistaken for separate month-to-month prices.

The source code is publicly inspectable but is not currently MIT/open-source. Screenpipe changed to its Commercial License in June 2026: personal non-commercial, nonprofit, educational and research use is free, while commercial source use beyond a seven-day organizational evaluation requires a paid license.⁵ Earlier MIT-licensed releases remain MIT, but buyers evaluating the current code should treat it as source-available with commercial restrictions.

Privacy depends heavily on deployment mode. Screenpipe’s current privacy policy says local-only recordings, transcripts and pipe data remain on the device; content is transmitted only when the user enables features such as sync, cloud AI, exports, connectors, team workflows or support.⁶ Optional cloud AI can route selected context to Google Cloud Vertex AI, Anthropic or OpenAI, while cloud transcription can use Deepgram, so “local-first” should not be read as “nothing can ever leave the machine.”⁶

The security architecture is unusually detailed for an early desktop AI product. First-party documentation describes a localhost-bound authenticated API, encrypted credential storage, app/window/content filters, local AI options and configurable enterprise policies.⁷ Source availability makes parts of those claims auditable, but the documentation is still produced by Screenpipe and is not a substitute for an independent security assessment of a particular deployment. Continuous capture also creates a large local concentration of sensitive material even when no cloud service is enabled.

Y Combinator lists screenpipe as an active Summer 2026 company and describes the same local/source-available context-capture product.⁸ Independent user-review evidence is much thinner: G2 currently has an exact Screenpipe profile but reports zero reviews, so it does not provide a meaningful satisfaction or reliability signal.⁹ No independent numeric evidence is used to infer product quality here.

screenpipe is best suited to developers and knowledge workers who want durable context across meetings, coding, research and everyday desktop work, and to teams willing to define explicit capture and access policies. MCP, local API access and scheduled workflows make it particularly relevant when existing agents need memory rather than another standalone chat interface.¹³

It is a weaker fit where continuous screen or audio capture is unacceptable, where third-party confidentiality rules prevent broad local recording, or where an organization cannot establish consent, exclusion, retention and access controls. Users who need a permissive current open-source license should also note the Commercial License before building screenpipe into a product or production workflow.⁵¹⁰