Atomic Agent Files

This release is mostly quality of life for cloud and local model setups, plus a privacy fix in the Telegram channel.

Fixed

• The Telegram progress bubble echoed raw tool output into chat history. The status line forwarded the tail of a tool's output, so a step that read a config file or grepped a repo could put a path, a match or a key fragment into Telegram history, which is forwardable and outlives the session. Labels are now built from stable identifiers only. The bubble is also sent silently (every message used to produce two push notifications), edits are throttled under Telegram's per-chat limit and back off on a 429 so they can no longer delay the final reply, and the whole indicator can be switched off with telegram.progressIndicator (#56, fixes [#58]).

• The managed llama-server kept holding RAM and VRAM after the terminal window was closed. Closing the window sends SIGHUP, which had no handler, so the process died before teardown and the daemon was orphaned until a manual models stop. It is now stopped on exit, configurable via localModels.managed.stopOnExit (default true) for anyone who prefers a warm model between sessions. A session registry makes sure only the last live session stops the daemon, which also fixes a pre-existing bug where the first of two open TUI windows to exit killed the model under the second (#57, fixes [#52]).

• Re-running the provider wizard could silently reset a custom endpoint. The base URL step opened on an empty field, so pressing Enter through it fell back to https://api.openai.com and clobbered a vLLM or LM Studio URL. The stored value is now prefilled (#63).

Added

• A reopenable cloud model picker. After first-time setup the server's model list used to be unreachable: the Cloud pane showed only the selected model and /model just jumped to the LLM tab. Enter on the model row and /model now reopen the provider's /v1/models list with the current model marked, and switching applies mid-session without re-entering the API key or base URL (#64, closes [#62]).

Updating

macOS / Linux

curl -fsSL https://atomicagent.io/install | sh

Windows (PowerShell)

irm https://atomicagent.io/install.ps1 | iex

Existing installs update in place; the TUI prompts for a restart afterwards.

Thanks @sosidudku1 for the fixes and the review that caught the Telegram privacy leak, @Ooooze for the reviews and the release, and @bxxvii and @nicopara on Discord for the reports behind two of these.

Full Changelog: https://github.com/AtomicBot-ai/atomic-agent/compare/v0.1.71...v0.1.72