Langfuse joins the lineup. Connect a self-hosted Langfuse and see your LLM apps from your phone: today’s calls, spend and errors, recent observations, and each trace down to its input and output. Sessions group traces into conversations you can read back. The Services tab now groups connections into sections you can collapse: Servers & containers, AI & models, Monitoring, Storage, Code and Network. Also fixed: a Gitea run’s job list no longer hides jobs added on a branch, recent Hermes sessions sort newest first, and chat recovers cleanly when a conversation fails to start.

peekstack is now free. Connect two services and use the whole app: every integration, the terminal, the widgets, search, syncing and demo mode. The only limit is how many services you connect at once, and you can swap which two whenever you like. If you run more than two, peekstack Pro lifts that limit with a one-time purchase, and adds multiple dashboards and the alternate app icons. No subscription and nothing to renew.

New: Hermes Agent. If you run the self-hosted AI agent by Nous Research, peekstack now watches it and works with it from your phone. See what it has been doing: health, sessions, an activity calendar and what it cost you today. Open any session and carry on typing, with the whole thread there, thinking and all. Its kanban board is in your pocket too. Read a task with its comment thread, answer with a comment of your own, move cards with a long press, and create one from the phone for a running dispatcher to pick up. A blocked task raises an alert, because a blocked task is the agent waiting on you. Explore what the agent is made of: every skill it has, with the instructions each one follows; the jobs it runs on a timer, which you can pause, resume or run now; and the files it can reach. And when the agent asks permission to run a command, the chat asks you. The command is shown exactly as it would run, with the agent's own choices, and it waits until you answer. Also new: a LiteLLM proxy can be managed from the app, creating and editing its models, teams and virtual keys, with a Test route button that sends a real request through what you typed before anything is saved.

Read the code, not just the commit list. Tap a commit to see exactly what it changed, line by line, and open a pull request to read the whole thing before you approve it. Browse a repository's files on any branch, with syntax highlighting. On iPad the file list stays where it is while the file you picked opens beside it. A very large change is shown as far as it will load and then says so, naming how many files it managed, rather than presenting part of a diff as the whole of it. Also fixed: a service sitting behind a single sign-on login now says so, instead of connecting and showing nothing.

Chat with your own models. • Talk to any model on your Ollama server, right inside the service. Replies stream in as they are written, with rendered markdown and highlighted code blocks you can copy in one tap. Reasoning models show their thinking as it happens, folded away once the answer starts. Models that take images take photos from your library. Every reply signs off with how fast it ran, and each conversation is waiting where you left it. • The same chat works with any OpenAI-compatible server: LiteLLM, vllm, llama.cpp, LM Studio. Add the address, paste an API key if it wants one, and talk. Gitea Actions, all the way down. • A repository's runs now open into the jobs of that run, and a job into its live step-by-step log. Steps tick green as the runner finishes them, the running one follows its own output, and the jobs still waiting for a runner are listed before they start. Also in this release. • Your contribution heatmap from Gitea, on its own screen and as a dashboard card. • Home settles as it loads instead of shifting under your finger, and its cards read clearly in light mode again. • Pull to refresh puts its spinner where you can see it.

Two new services: TrueNAS and Prometheus. Pools, capacity, scrubs and alerts from your NAS; what is firing right now, which scrape targets stopped answering, and which rules fail to evaluate. Home-screen widgets. One answers “is anything wrong” at a glance. The other shows a service of your choosing with the same body as its card in the app: pools with their capacity bars, monitors with their beat strips, alerts by name. Both refresh in the background and carry an “as of” time, so they never pretend to be newer than they are. The overview also comes in a lock-screen size, and a running model download shows its progress in the Dynamic Island. Choose your app icon: Ink, Parchment or Phosphor, with a matching green terminal theme. Open a shell inside a Docker container from Portainer, in the same terminal layer as everything else. Pinch to zoom in a live terminal. The TLS-off marker is now a small red shield rather than a large pill, and cards no longer fade into the page in light mode. Search no longer flickers while you type.

Two new services: Ollama and ComfyUI. See which models are loaded and the memory they hold, pull or remove models, and watch what the GPU is rendering with a live preview. iPad gets a real sidebar: always visible, collapsible to an icon rail. Background refreshes now land silently instead of pulling the screen down. A streaming terminal no longer slows the rest of the app. Stopping a model download now stops just that download. Polished wording throughout the app.