Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.ticket-rs.io/llms.txt

Use this file to discover all available pages before exploring further.

Henrik Albihn
Henrik Albihn
Creator of ticket-rs

​
Someone Else Said The Quiet Part Out Loud

At AI Engineer Europe (presented by Google DeepMind), Mario Zechner — creator of pi, the core agent framework that powers openclaw — put this on screen during his talk, Building Pi in a World of Slop:
Mario Zechner presenting at AI Engineer Europe, with the highlighted quote: 'You installed Beads, completely oblivious to the fact that it's basically uninstallable malware.'
“You installed Beads, completely oblivious to the fact that it’s basically uninstallable malware.”
That’s a spicy framing. But he’s not wrong about the underlying mechanic, and it’s the exact failure mode ticket-rs was built to avoid.

​
The Daemon Tax

Steve Yegge’s beads introduced a lot of good ideas — structured, terminal-native issue tracking designed for coding agents, dependency-aware triage, graph-native prioritization. I wrote about this already: beads was one of the first projects to ask the right question. Where it gets into trouble is the runtime. beads runs a long-lived SQLite-backed daemon. That daemon:
  • Persists state outside your repo, in a process you didn’t ask to supervise
  • Holds file locks that outlive your terminal session
  • Has to be cleanly shut down or you get zombie state
  • Adds IPC overhead to every single command your agent runs
For a human using the CLI once every few minutes, that cost is invisible. For a coding agent running triage and ready and next in a tight loop — dozens of tool calls per session — you pay that tax on every single hop. And when the daemon misbehaves, you’re debugging a background process instead of your code. Mario’s phrasing is harsh, but the point lands: software that installs a persistent background process without making that contract obvious is software that’s hard to reason about, hard to uninstall, and hard to trust inside an autonomous loop.

​
Why ticket-rs Is Stateless By Design

ticket-rs takes the beads ideas — graph-native backlog, dependency-aware triage, CLI-first UX — and deletes the daemon.
  • No background process. tk is a single ~9MB Rust binary. It starts, reads files, exits.
  • State lives in your repo. Issues are markdown files with YAML frontmatter in .tickets/. git log is your audit trail.
  • The graph is incremental. Analytics results (PageRank, betweenness, critical path) are cached to .tickets/.cache/ and invalidated per-file by SHA-256 hash. Change one ticket, recompute the delta, not the world.
  • Every invocation is fresh. No zombie locks, no “is the daemon running,” no supervision burden for you or your agent.
The measurable outcome: tk triage completes in 14–21ms on a 500-ticket repo. A daemon-backed system pays an IPC round-trip before it even starts work. Zero daemons turns out to be faster, not just cleaner.

​
Agents Don’t Want Daemons Either

This is the deeper point, and it’s what Mario is really getting at with pi. A coding agent is a stateless function call. Context in, next step out. Everything in its world should be composable, inspectable, and cheap to call. The moment you introduce a long-running process into that loop, you’ve added:
  1. A supervision contract the agent can’t fulfill. It can’t systemctl status your daemon from inside a sandbox.
  2. A hidden failure mode. When the daemon is wedged, every tool call looks like a tool bug.
  3. A trust problem. “What else is this thing doing on my machine?”
ticket-rs — and pi — converge on the same answer: small, fast, stateless CLI primitives that agents can call without inheriting a runtime. That’s why ticket-rs ships as a pi package out of the box — the model is the same all the way down.

​
The Thesis, Restated

  • Daemons are a liability when your primary user is an autonomous loop.
  • The state you want is in your git repo, not in a SQLite file a sidecar process owns.
  • The graph you want is derived from that state, incrementally, every call.
  • Fast, stateless, inspectable tools compose. Long-running services don’t.
Steve deserves real credit for naming the problem. ticket-rs is what the solution looks like when you delete the daemon.

Install ticket-rs

60 seconds. One binary. No daemon.

Watch the talk

Mario Zechner — Building Pi in a World of Slop