PrivateGPT vs screenpipe

PrivateGPT 41

The last release is 1.0.1, dated 16 June 2026 in the changelog and tagged 18 June, 141 days before the check (10 of 30). No release or dated changelog entry in the 90 days since 8 August (0 of 20). Maintainers commit most days and outside pull requests merged in October, but 23 pull requests are open and the stale bot closes issues, including two security reports, with no maintainer reply (13 of 25). Homebrew tap, wheel index and Docker images ship with releases, and Anthropic SDKs work against it. No product MCP server of its own, though it connects to remote MCP servers (10 of 15). Dependabot, uv.lock, actions pinned by commit since 19 October and dependency CVE upgrades in June, July and August (8 of 10). 10 + 0 + 13 + 10 + 8 = 41.

screenpipe 82

app-v2.7.84 on 1 October 2026 and mcp-v0.20.2 on 27 September (30). 81 app tags and 7 MCP tags since 5 July (20). Fixes land daily, but issues close automatically after 14 days without activity and pull requests after 7, first-time contributors' pull requests are closed on arrival with a stock note, and we couldn't see who answers or how fast (10 of 25). In the official MCP registry as io.github.screenpipe/screenpipe-mcp under a GitHub-verified namespace, but the registry's latest is 0.19.4 of 29 August against 0.20.2 on npm (13 of 15). CI passing, with Dependabot and gitleaks (9 of 10).

PrivateGPT 60

Read with the self-hosted rule. PrivateGPT is Apache-2.0 with nothing to buy. Zylon, the company's enterprise platform built on it, is a separate product sold through demos, and is not the surface graded. No payment protocol (0). Free with no account and no card, so 20, 20 and 20 on the last three lines.

screenpipe 30

Scored on the paid app, as the rubric asks for software with a paid version and as the Marmot dossier scored its hosted plan. No payment protocol (0). Free, Basic at $21 a month and Business at $42 a seat a month on a public pricing page, plan prices with no per-call unit (10). A $0 plan exists, but the page doesn't say whether it needs a card, the desktop app records only for a signed-in account, Free history reads stop at the last 24 hours in the source, and source and CLI builds are free only for non-commercial use or a seven-day evaluation at a company. The Marmot dossier gave 10 for a free option hedged this way (10 of 20). The CLI records with no account and screenpipe auth token prints the local key, so an agent on the machine can get access, but buying Basic or Business takes a person in a browser (10 of 20).

PrivateGPT 32

Read with the tool checklist. Authentication is off by default, and when on it is one static secret compared with the whole Authorization header, with a sample secret:key value in settings.yaml. There are no per-user keys or scopes, private-gpt serve binds 0.0.0.0 by default and CORS allows every origin, so a default start answers anyone on the network. The docs disagree on the header (Bearer, Basic or an Anthropic key) (6 of 30). Tools are chosen per request and code execution has internet off by default, but bash runs as a host subprocess with resource limits, not in a container, and nothing asks for approval (6 of 20). Web fetch, web search, documents and MCP servers bring untrusted text to the model. A public prompt-injection report (#2270) was closed by the stale bot, and no guidance was found beyond a 19 October change that bounds tool output (3 of 15). OpenTelemetry traces cover every request, model call, tool and sandbox step, off by default, with local JSON lines exporters (10 of 15). SECURITY.md since 23 July 2026 asks for private reports through GitHub or Discord. No bounty, no certification, and the June and July security reports in issues were closed by a bot with no reply (7 of 20). 6 + 6 + 3 + 10 + 7 = 32.

screenpipe 48

Graded on the MCP server and the local API it calls. The API asks for a Bearer key on every request by default, localhost included, and forces auth on when it listens on the LAN. The key is sp- plus 8 hexadecimal characters (about 4.3 billion values), generated on first run, kept in the local secret store and regenerable, and one key reaches every route, raw SQL and pipe creation included. Pipes get their own sp_pipe_ tokens limited by their permissions, but the MCP server and any outside agent use the main key. That's 20 for a revocable key, less 10 because the getting-started page documents passing it as a token query parameter (10 of 30). Per-pipe allow and deny rules for apps, windows, content types, hours and endpoints, but no read-only mode for the main key, and create-pipe, run-pipe, control-recording and merge-speakers run without a confirmation step (10 of 20). Results carry screen text, transcripts and messages written by anyone. The bundled skills the MCP server serves through screenpipe-skills tell the model to treat captured content as untrusted evidence and ignore commands in it, though the tool descriptions and the MCP docs page don't. One shipped pipe template (speaker-reconciliation, off by default) carries the vendor's own instruction to AI agents to add Screenpipe's header to every file they create or edit, even outside the repository (8 of 15). Pipe logs through pipe-logs. MCP calls carry x-screenpipe-client and x-screenpipe-agent headers, which feed usage attribution rather than an audit view, and we found no per-call audit log (6 of 15). security.txt valid to 30 June 2027, a disclosure policy, a trust centre and a SOC 2 Type 2 report under NDA, with ISO 27001 in progress and GitHub private reporting on. No SECURITY.md, no published advisories and no bounty (14 of 20).

PrivateGPT 64

Read for an HTTP API an agent calls on a local instance. Token counting, top_k style retrieval settings, citation output and response byte caps let a caller size context (18 of 25). Artifact listing by namespace and retrieval filters, with no documented cursor pagination (12 of 20). FastAPI validation errors name the field and the validate route checks a request before running it, but there is no error catalogue (12 of 20). Async messages and ingestion return task ids with status, cancel and delete routes. No idempotency keys were found (10 of 20). Anthropic SDKs and Claude Code work by pointing ANTHROPIC_BASE_URL at the server, but a model server and an embedding server have to be configured first (12 of 15). 18 + 12 + 12 + 10 + 12 = 64.

screenpipe 75

Graded on screenpipe-mcp. 33 tools by default, more than 30 (5). The four team tools load only with a team token, and screenpipe-mcp-http registers a single read-only search tool, which we count as a read-only subset (5 back). The 33 definitions hold about 22,500 characters of strings, about 33,000 as JSON, roughly 5,600 to 8,300 tokens before any call depending on the count (10 of 25). limit and offset, time ranges, app, window, speaker and tag filters, per-result truncation at 1,000 characters by default and screenshots off unless asked for (20). Failed backend calls return isError with the cause and the next step for 400, 401, 403, 404 and 5xx, and a message when the app isn't running, but several missing-argument errors come back as plain text without isError (16 of 20). Every tool carries annotations, 21 with readOnlyHint true, merge-speakers marked destructive, and create-pipe and run-pipe marked non-idempotent and left at the spec's destructive default. No idempotency keys (18 of 20). 17 tools need no arguments and the defaults are sensible, but the official SDK is JavaScript only (11 of 15).

PrivateGPT 64

Apache-2.0 (30). The OpenAPI description says no data leaves the execution environment, but the docs list cloud gateways as model providers and web search goes to outside services when enabled, and no privacy statement covers the software. Zylon's privacy policy covers its website (15 of 30). The 1.0 rewrite was marked as a breaking change in the changelog, and there is no deprecation policy (6 of 20). No telemetry to Zylon found in the source. OpenTelemetry tracing is off by default and its exporters write local files or to the owner's collector, documented on the observability page (16 of 20). 30 + 15 + 6 + 16 = 67.

screenpipe 72

The Screenpipe Commercial License publishes the source for reading and modification, with free use limited to personal, non-commercial, non-profit, educational and research purposes and a seven-day evaluation at an organisation, and it covers the npm packages, screenpipe-mcp included. Clear terms and auditable source, not OSI, and versions released earlier under MIT stay MIT (18 of 30). The privacy policy of 24 September 2026 names its third parties and deletes server-side data within 30 days, and it and the security page say data leaves the device only through settings you turn on, while the README says PostHog analytics (a stable installation ID, and email once signed in) and Sentry are on by default. Since 17 September 2026 remote support logs are also on by default for signed-in users, while the privacy data-flow page says log bundles leave only when you send them (18 of 30). No deprecation policy, but the weekly changelog names removed settings and functions and keeps a deprecated alias, and existing lifetime licences stay valid while new ones aren't sold (10 of 20). Telemetry is on by default and disclosed in the README since 15 July 2026, with a switch in Settings > Privacy > Analytics, and the MCP package's README describes its Sentry reporting with four environment variables to turn it off (12 of 20).

PrivateGPT 70

Read with the local-software lines, since PrivateGPT runs on the owner's hardware with no hosted service behind it. Homebrew tap, uv tool install from Zylon's own wheel index (not PyPI) and Docker Hub images zylonai/private-gpt 1.0.1 of 18 June 2026. Python 3.11 only, stated in pyproject.toml, and it needs a separate OpenAI-compatible model server (17 of 20). tests.yml runs format, lint, typecheck and the tests affected by each push on main, with the full suite of 217 test files on release pull requests. We couldn't see run results because the GitHub API refused us, and issue #2441 of 1 November reports the Claude spec update workflow failing (15 of 25). GitHub shows 3 open issues of about 1,250, because a stale bot closes quiet issues, two security reports among them, with no maintainer reply (12 of 25). release-please keeps a changelog with breaking changes marked, but no release has been cut since 1.0.1 while main carries 51 commits since 8 August, among them a breaking observability change of 23 October (11 of 15). Version 1.0.1 (15). 17 + 15 + 12 + 11 + 15 = 70.

screenpipe 65

Read with the local-software lines, as the Goose and Aider calibration dossiers do, though Screenpipe also sells plans. Desktop builds for macOS (Apple Silicon and Intel), Windows 10 and 11 and a Linux x86_64 AppImage, the screenpipe CLI on npm, and screenpipe-mcp on npm (20). The Rust CI workflow passed on every main run we saw on 3 October, with MCP, frontend, end-to-end and Windows workflows beside it (25). 9 open issues, three of them crash or capture reports from the last fortnight (#7352 timeline worker panics on macOS, #7190 SIGABRT on the Linux AppImage, #7262 the Google Meet screen-share picker), and a daily workflow closes issues after 14 days without activity, so the open count says little (14 of 25). 81 app tags and 7 MCP tags since 5 July. The weekly changelog names removals and keeps a deprecated alias (ocr_text) but has no breaking-change sections, and its last entry is the week of 7 September while tags ran to 1 October. The README says main moves fast and breaks things (6 of 15). The interface an agent installs, screenpipe-mcp, is 0.20.2 with no stability statement, and the app's 2.7 doesn't change that. The line counts one way or the other (0 of 15).