AI Agent Life, from Zero · part 13
[Agent 101 #13] See what your fleet of AI agents is doing — from your phone: Muninn adds a Kanban board
❯ cat --toc
TL;DR
Hermes ships with a Kanban — agents open, claim, and report on cards there. But away from your desk, all your phone gets is Telegram's plain text, and the board's shape disappears into a wall of lines. Muninn now pulls that board onto the phone: horizontally-scrolling Running / Blocked / Done columns — who's working on what, which card got blocked, which one failed — at a glance. Zero backend changes, pure client, over the same encrypted tunnel. Read-only for now.
The short version: you've got a fleet of agents, but all you see is the text they spit out
Earlier in this series you got Hermes running and split it into several sibs (Part 8) — each with its own brain, each off doing its own thing.
The problem: you can't really see what they're doing.
Hermes has a built-in Kanban: what's queued, what's running, what's blocked, what's done — the agents log all of it. Open the dashboard in a browser on your computer and it's a clean, draggable board.
On mobile, though, you're back in Telegram — and Telegram only gives you text. A whole board flattened into lines — no columns, no color, "who's doing what" is something you reconstruct by eye. What you lose is the at-a-glance.
That's the gap Muninn closes: it brings the same board onto the phone as an actual board.

Three columns: who's working, who's blocked, what's done
The board scrolls sideways, one status per column, each with its own color. The three I look at most:
Running (green) — who's actually working. A running card has a spinner in the top-right showing which sib is on it and how long it's been running. In the shot above, hikari is "building a Shopee price monitor for the ASUS Ascent GX10," 27 seconds in. No asking, no scrolling back through messages — open the app and you see it.
Blocked (red) — who got stopped, and why. This is the interesting column. Several of these are cards the agent blocked itself: a task wanted it to edit ~/.hermes/config.yaml and restart the gateway — control-plane stuff it judged out of bounds, so it refused.
So the board doesn't just show you progress, it shows you whether they're coloring outside the lines. Clicking down the red column is basically auditing whether your fleet of AIs is staying inside its guardrails.
Done (gray) — finished work stacks up here. 44 of 50 cards sit in Done; tap any one for its full output. One scroll down the column tells you what the sibs got through all day.
Every card carries the essentials: priority, the sib on it, how long it's sat, a two-line preview; tap in for the full content, the comment thread, and each run's log and failure messages.
How it works: zero backend changes
This is the part I'm happiest with — the whole feature is client-side on the phone; the home machine didn't need a single code change.
Hermes's dashboard already exposes the board as a REST API (/api/plugins/kanban/*). And Muninn already has that encrypted iroh P2P tunnel to the dashboard (the bridge from Part 12). So all the phone has to do is reuse the same connection that already carries voice and files, call one more API to read the board, and render it as columns in the app.
- No new server, no bridge change, no dashboard change.
- The board follows whichever agent you're connected to (one agent, one board).
- It tells whether the board changed by comparing a single
latest_event_id, so it can auto-refresh cheaply.
The whole thing stands on the groundwork the earlier posts laid down — it's one more layer of UI on top.
Honest note: this version is read-only
This version is read-only. I wanted the read-only view solid first — columns, cards, tap-in detail, auto-refresh.
Dragging cards to change status, commenting, opening new cards — those write actions aren't in yet; they're the next phase. Not because it can't (same REST path), but because I wanted the "seeing" to feel right before piling more on.
So if you're in the "I just want to glance at what my fleet is up to while I'm out" camp — this version is enough. Want to drive it hands-on? Hang tight.
The board is the natural extension of "see your agents from anywhere." Prereqs:
FAQ
- Does Muninn's board need a separate backend?
- No. Hermes already ships a Kanban (agents open, claim, and report on cards there). Muninn just reads that same board over the existing iroh P2P tunnel and draws it on your phone. No new server, no account, no cloud.
- Can I drag cards or change status from the phone?
- Not in this version — it's read-only. You can see every column, tap into a card for its full content, comment thread, run logs, and failure reasons. Dragging, commenting, and creating cards are write features for the next phase.
- Whose board am I looking at?
- Whichever agent your phone is connected to right now. Each Hermes agent runs its own dashboard and its own kanban.db, so when you switch sibs the board switches with your connection.
Read next
- 2026-06-23[Agent 101 #12] Reach your home AI agent from anywhere: Muninn, a private iOS app over iroh P2P
Hermes runs at home, but you lose it the second you walk out. Bridging through Telegram works but it's fiddly and routes every message through someone else's server. Muninn is an iOS app built for Hermes: give your agent one command, scan a QR, and your phone connects straight home over an encrypted iroh tunnel — no cloud in the path.
- 2026-07-01[Agent 101 #15] Hermes /learn: I had a local 27B write its own reusable skill
Hermes has a /learn command that turns 'something you just did' into a reusable skill — a SKILL.md. I wired it into my own fleet: one Kanban card, a local 27B running on a modded 2080 Ti, and about 3 minutes later it handed back a clean, spec-compliant skill — plus two implementation details the docs don't spell out (slash command vs. dispatch, and where skills actually live). A plain-language walkthrough of what /learn does, how to use it, and where its limits are.
- 2026-06-29[Agent 101 #14] One spec, three assistants, three Tetris games: a Hermes Kanban dispatch test
After raising a fleet of assistants, I gave them the same one-line 'make a Tetris game' spec — no details at all — one card each, and let them each write a web Tetris in a single shot. I touched zero lines of game code; I only published the result. The surprise: from that one line, the Hermes harness plus a local model I tuned myself (on a modded 2080 Ti) filled in things I never asked for — a ghost piece and wall-kick — in one shot. You can play all three.
- 2026-07-21[Agent 101 #16] Hermes config health check: 5 silent gotchas that make your assistant act weird
Part 11 said a haywire assistant is usually a broken car (tools, config, memory), not a dumb engine (the model). This is that checklist: a context_length set at the wrong level silently compresses early, Qwen thinking left on runs 10x slower, an MCP tool that connects but every call fails, and a sib running a different model than you think. Five real config gotchas, each with a check you can hand to your agent to run on itself, plus the fix.
Don't miss the next one
Subscribe, and you won't.
One-click unsubscribe anytime.