Show HN: Yet another agent manager (but we needed it) (varinomics.com)

1 points by imakris ↗ HN
Hello HN,

I am Ioannis and I would like to introduce Logonomic, a desktop application we made to help us work with LLM CLIs (e.g. codex, claude, opencode etc) in a more organized way. It runs the CLIs already installed, with their existing accounts and configuration and organizes sessions by repository. It is available for Windows and Linux, with a preview for macOS. There is also a mobile companion app, to work remotely (Android and soon iOS too).

We built it originally to assist us in our own work, to address the common case of multiple agents running concurrently in many repositories. It was hard to keep track of what each one is doing and which one needs attention or requires action.

There are many tools by now broadly that broadly target the same problem. This one is a native C++/Qt application with our own terminal emulator. Some of its features:

- An 'overview' mode that shows all running sessions, optionally grouped by repository, each with its status, title and summary that the agents update regularly.

- Agents add a trailer to their commits, enabling to re-open the conversations that produced them.

- Terminals run in their own processes and survive UI crashes.

- A plot that shows line counts over time, by directory or language, where clicking on a sample brings up the corresponding commit in the git log.

- A treemap that shows how the repo's files are organized, with a diff mode that colours changed files, whose diff opens with a double click.

- Token usage and remaining account quotas shown where available, with usage broken down by day and repository.

- Sessions can exchange messages (1:1 or broadcast), may hand tasks to each other or continue other session's work.

The supported CLIs are Codex, Claude Code, OpenCode, Antigravity, Grok Build, Kimi Code and Qwen Code.

The free edition has no time limit and runs up to four terminals. The paid editions have perpetual one-time licences from €150 for 2 seats and include 12 months of feature updates, no terminal limit, GitHub/GitLab CI integrations, as well as an Android companion app (iOS will follow).

The application itself is proprietary. Many of its core components are open source, among them the terminal emulator and the plotting library: https://github.com/Varinomics

This is the first public release, so there will be rough edges, and we would like to hear about them.

7 comments

[ 4.8 ms ] story [ 18.7 ms ] thread
Does it support oh-my-pi? And how well does it handle projects with multiple repositories? like frontend, backend?
You can definitely run omp it inside logonomic, but there is no dedicated integration for it yet. CLIs running inside Logonomic are heavily hooked to report all sorts of activity, usage, conversation history etc. If OMP allows integration, even to a certain degree, we can write an add-in for it. I will look into it. Multiple repos are the intended use case. I don't know if I understand your question correctly, I assume you refer to an application whose frontends and backends live in separate repositories. You can have them on the same window, with sessions on both, or work isolated when necessary. Going from one mode to the other is usually one click. Sessions working on both can also see and message each other, e.g. to coordinate API changes between backend and frontend.
To lookbk, whose comment were flagged, but the question was legitimate: Logonomic shows you the worktrees each session touched and whether they are done working in each. But it will not warn you if multiple sessions are working on the same worktree. I did not think it should, becasuse it is probably not too uncommon to have a large monorepo project with many agents working independently on different parts. But it might actually be a good idea for a feature to warn, optionally, whoever does not want that to happen (myself included, since I prefer to work with smaller repos and multiple worktrees), so I am probably stealing it (thanks!).
Really interesting approach, looks slick! I am still trying to land on a choice for a top-down view of agentic work. One question: does this work with setups like OpenRig and their workshop pattern? I like the idea of interacting with a lead agent, but having visibility on whether the builder agents are on/off the rails is something I haven’t found the “right” UI/UX for yet (rig tui doesn’t feel quite right).
Logonomic's orchestration infrastructure is not as elaborate as openrig's. It can do lead and workers setuos, but the planning would happen in the conversation, not in a separate planning engine. For example, it is feasible to instruct a session to investigate a refactoring, then delegate implementation/review tasks, coordinate with other sessions etc. There is infrastructure for cross-launcher delegation, wakeups and execution visibility. Subagents would normally use the CLI's native mechanisms, unless explicitly requested otherwise.

However, it cannot yet do persistent, application-managed planning with dependencies and revisions.

This kind of orchestration was, in fact, an idea that we worked quite a bit on earlier, but we couldn't use it meaningfully ourselves, thus was eventually defered.

We do have a couple of features under development going this direction, one of which is (possibly) described in your question: supervisors which sample, audit and intervene, if the worker seems to drift too much, triggered by certain heuristics. But the main debate for each feature combining execution visibility with extended orchestration, is whether reduces the effort required for sound decisions, or would eventually be asking users to supervise more than they can meaningfully understand.

Now, despite all that, it is probably feasible to get openrig to run inside logonomic, since openrig does have sufficient infrastructure to allow this kind of integration, so I think it would be a good next feature, for anyone who might like to work with openrig.