We've introduced a new workflow_dispatch action that allows for an immediate, safe rollback of production to any previous Docker image version. The workflow handles version normalization and ensures system state is preserved by retagging the target image as :latest and restarting the containers, all while sharing a concurrency group with standard deployments to prevent race conditions. If your production deploy goes sideways, you can now quickly restore a previous version via the GitHub Actions UI.

This release adds interactive inline keyboards to Telegram, allowing you to quickly mark items as read, archive notes, or complete reminders with a single tap. These actions execute immediately against your storage without needing an LLM round-trip. It's a faster, smoother way to manage your queue and tasks directly from your chat.

We've integrated Telegram inline keyboards to streamline interactions with your reading queue, notes, and reminders. Instead of requiring a full LLM round-trip, tapping a button now executes actions like archiving or marking items as complete directly against your local storage. This update significantly improves UI responsiveness and makes managing your lists feel much more native and efficient.

This release introduces five targeted security improvements, including enhanced URL validation to block cloud metadata endpoints and CGNAT, plus a bounded rate-limiter to prevent memory DoS. We've also added strict payload size validation for all text inputs, expanded log redaction for sensitive fields, and replaced bearer-token comparisons with a timing-safe alternative. These changes significantly reduce the attack surface for exposed endpoints and untrusted content inputs.

We've deployed a comprehensive security update to bolster the application against several common attack vectors. This includes SSRF prevention via stricter URL validation, protection against memory exhaustion by implementing rate limit eviction, improved secret redaction in logs, enforced content size limits to prevent abuse, and timing-safe token comparison to mitigate side-channel leaks. These updates ensure a more robust and secure architecture overall.

This update enhances how the agent handles previously saved conversations by adding explicit instructions to its system prompt. The focus shifts towards using semantic search via search_knowledge as the primary method for retrieval, ensuring that older conversations stored as notes are correctly indexed and discoverable. These changes improve the reliability and accuracy of the assistant when referencing past interactions.

This update registers the save_conversation tool as a core agent capability, enabling the agent to better capture significant personal content. We've added specific system prompt guidance that mandates proactive user confirmation before saving and instructs the LLM to prioritize semantic content search over brittle type-based lookups for retrieval. This improves long-term memory reliability by ensuring conversations are indexed correctly and easily discoverable via search_knowledge.