From 868bdff078c647eee2bc18fe2e0ba91ade65038e Mon Sep 17 00:00:00 2001 From: Adeel Ahmad Date: Sun, 11 Oct 2026 00:41:00 +0000 Subject: [PATCH] docs: list the prompt-injection layers and the setup budget as built "Stay private and safe" now covers flagging instruction-like text (patterns or the optional local classifier, #329, #347), asking before the browser acts on what the assistant wrote (#333), breaking up links in chat-room answers (#338), the archive as data over MCP (#343) and the audit log of what was caught (#344). "See what it costs" adds the overall budget asked for in setup (#281). "In progress" drops what merged and adds #337, #341 and #346. Co-Authored-By: Claude Opus 5.5 (1M context) Claude-Session: https://claude.ai/code/session_01HDx7PPUauyjMn1LZs8JAsS --- README.md | 34 +++++++++++++++++++--------------- 1 file changed, 19 insertions(+), 15 deletions(-) diff --git a/README.md b/README.md index 136bddcab..33ed1357f 100644 --- a/README.md +++ b/README.md @@ -96,9 +96,11 @@ Everything below is built and merged. Work that is still in a pull request is li notifications to webhooks, Slack, Discord and Matterbridge. ([Processing](docs/processing.md#workflows-and-the-canvas), [notifications](docs/notifications.md)) - **See what it costs.** An activity ledger of every model call, with reports, charts and budgets that cap spending. - A meter in the top bar shows spend against the budget, and the Costs page breaks it down by day, model and what - cost most. Routine choices go to small decision models, and the LLM only when it adds value. - ([Activity and costs](docs/activity.md), [budgets](docs/budgets.md#the-meter-and-the-costs-page)) + Setup asks for one overall budget after a cloud model, with a soft limit that warns and a hard limit that stops paid + calls. A meter in the top bar shows spend against the budget, and the Costs page breaks it down by day, model and + what cost most. Routine choices go to small decision models, and the LLM only when it adds value. + ([Activity and costs](docs/activity.md), [budgets](docs/budgets.md#one-budget-for-everything), + [the meter](docs/budgets.md#the-meter-and-the-costs-page)) - **Keep it up to date and backed up.** Lens checks for a new release once a day and tells admins by email or a notification, and upgrades itself through a small agent on the machine, from the API; nothing installs unless an admin says so. Settings › Backups takes, lists, downloads and restores backups, takes them on any schedule, and can @@ -109,11 +111,14 @@ Everything below is built and merged. Work that is still in a pull request is li Grafana next to Lens, already connected, with the Lens dashboard as Grafana's home. ([Telemetry](docs/telemetry.md), [Prometheus](docs/prometheus.md)) - **Stay private and safe.** Passkeys instead of passwords, encryption at rest with a key per namespace, roles per - namespace, and an audit log. What the assistant reads (recordings, documents, web pages, tool results) reaches the - model marked as data, not instructions, and once an answer has read such text, sending anything out waits for your - yes. Serve Lens over HTTPS with certificates from Let's Encrypt or ZeroSSL that - renew on their own, or reach it from anywhere through a Cloudflare tunnel with nothing opened on your router. - ([Encryption](docs/encryption.md), [access](docs/access.md), + namespace, and an audit log. Prompt injection meets several layers: what the assistant reads (recordings, + documents, web pages, tool results) reaches the model marked as data, not instructions, and text that reads like + instructions to an AI is flagged (by free patterns, or an optional classifier model that runs on your machine and is + fetched only when you pick it). Once an answer has read such text, sending anything out, or the browser opening or + typing what the assistant wrote, waits for your yes. Chat-room answers have links they built themselves broken up, + agents over MCP get the archive as data, and what Lens caught goes into the audit log. Serve Lens over HTTPS with + certificates from Let's Encrypt or ZeroSSL that renew on their own, or reach it from anywhere through a Cloudflare + tunnel with nothing opened on your router. ([Encryption](docs/encryption.md), [access](docs/access.md), [prompt injection](docs/assistant.md#third-party-content-and-prompt-injection), [HTTPS](docs/https.md), [remote access](docs/remote-access.md)) - **Use open standards.** RDF and Dublin Core for every record, OAuth for apps and OpenID Connect sign-in, MCP, MQTT, @@ -136,20 +141,19 @@ These are open pull requests, not yet part of Lens: ([#287](https://github.com/adeelahmad/lense/pull/287)), without Matterbridge; and the assistant's own email address, over IMAP and SMTP ([#314](https://github.com/adeelahmad/lense/pull/314)). - **Assistant safety:** running MCP servers that are programs (stdio) in a sandbox - ([#304](https://github.com/adeelahmad/lense/pull/304)); flagging third-party text that reads like instructions to an - AI ([#329](https://github.com/adeelahmad/lense/pull/329)); and asking before the browser opens or types what the - assistant wrote after reading a page ([#333](https://github.com/adeelahmad/lense/pull/333)). + ([#304](https://github.com/adeelahmad/lense/pull/304)). - **Cloud model presets** in Settings and the setup wizard ([#334](https://github.com/adeelahmad/lense/pull/334)). - **Search:** a Similar tab, and similar content for text selected anywhere ([#307](https://github.com/adeelahmad/lense/pull/307)); NSFW pictures classified by a local model, per namespace ([#306](https://github.com/adeelahmad/lense/pull/306)). -- **Budget in setup:** an overall AI budget with soft and hard limits, asked for in setup - ([#281](https://github.com/adeelahmad/lense/pull/281)). +- **Sources:** watching a local folder in four clicks ([#341](https://github.com/adeelahmad/lense/pull/341)). +- **On a phone:** the web app made usable on a phone ([#346](https://github.com/adeelahmad/lense/pull/346)). - **Running it:** an upgrade page with live steps that rolls back when the new version doesn't start ([#262](https://github.com/adeelahmad/lense/pull/262)), HTTPS set up in Settings ([#326](https://github.com/adeelahmad/lense/pull/326)), structured logs with a log viewer and syslog output - ([#284](https://github.com/adeelahmad/lense/pull/284)), and built-in usage analytics - ([#274](https://github.com/adeelahmad/lense/pull/274)). + ([#284](https://github.com/adeelahmad/lense/pull/284)), built-in usage analytics + ([#274](https://github.com/adeelahmad/lense/pull/274)), and turning Prometheus and Grafana on from the web app + ([#337](https://github.com/adeelahmad/lense/pull/337)). - **People:** email invitations, and removing someone for good ([#286](https://github.com/adeelahmad/lense/pull/286)). - **Lens.app for macOS:** a menu-bar app that carries microsandbox ([#280](https://github.com/adeelahmad/lense/pull/280)).