claudio

On screen

Read the banner, the transcript and the state line, and find past conversations.

claudio is meant to be heard, but its screen keeps a written record in step with the voice. This page covers claudio's own screen, in a terminal: how to read the banner, the transcript and the live state line, and where past conversations are saved.

Another front end shows the same record its own way: bb's panel shows the transcript as a chat, and its header shows what claudio is doing. The saved transcripts are the same whichever you use. Platforms lists the front ends.

Read the banner

When claudio starts, it clears the terminal and prints its mark, a small dithered orange sun, with its name ("claudio" unless you gave it another) and the line "Talk, and your sessions work." beside it. Under the mark are the settings in use and the keys you need, and once it's listening it says “Hi, how can I help you?”:

banner, under the mark
  orchestrator Sequencer
  brain default/low
  style plain
  voice orion
  echo cancel on
  memory on
  ears flux-general-en
  eager eot 0.3
  eot 0.7
  eot timeout ms 1500
  barge in on
  push to talk off
  address me off
  narrate handed-off
  mouth deepgram
  local speed 1.0
  local model compressed
  code colors on
  layout inline
  Enter cuts · Ctrl-C quits · claudio setup

Two of these facts change what you can do:

  • echo cancel off: claudio hears itself on speakers, so wear headphones to talk over it
  • memory fresh: this is a new conversation; memory on means the brain resumed the last one

With the ElevenLabs mouth, voice shows elevenlabs and the start of the voice id.

Read the transcript

Each line starts with a glyph that says who it's from:

GlyphLine
›you
◆claudio; one paragraph per thought, wrapped lines indented under it
▸a session reporting back, with the request it was for
→what the brain is reading or running
⚙a setting that changed
·a note: cut, replaced, muted, reconnecting
!a warning

Here you say "mm-hm" mid-reply, so the reply carries on, then a spoken command changes the voice. The ⚙ voice → thalia line is the setting changing:

› what changed on this branch?◆ Checking the log. → git log --oneline main..◆ Three commits: the migration, its tests and a fix. › mm-hm The fix is for the date column. › slash voice thalia· /voice thalia⚙ voice → thalia◆ Thalia from now on.

Backchannels show dim, as do turns that don't start with its name when address_me is on; the first of those in a run adds · not for me: start with claudio. A turn that does start with the name shows it as you said it, though the brain gets the turn without it. Three more rules keep the transcript readable:

  • Sentences appear as they're spoken: each sentence starts when its audio starts playing, and its words fade in from dim over the length of its audio. A long reply keeps its last few lines live; the lines above are already in the transcript
  • Tables and code appear whole, where the voice reaches them, and are never read aloud
  • Lines wrap at 90 columns, however wide the terminal
  • A dim rule with the time marks a gap of more than 5 minutes between your turns

With layout set to split (/layout split, or claudio setup), your turns sit on the right in italics, ending with ‹, instead of starting with ›, and claudio's stay on the left. Under 50 columns they fall back to › lines. The saved transcript is the same either way.

what's Sequencer doing? ‹◆ Sequencer is halfway through the migration tests.

Read the state line

The last line of the terminal, under a dim rule, updates four times a second. It starts at the bottom of the terminal and stays there as the transcript scrolls up above it. From left to right, it shows:

  1. A spinner, or a still ■ while muted
  2. The mic level, from blank to ▇
  3. What claudio is doing: warming up the brain… at launch, on the same line listening then takes, listening, thinking, speaking, paused, muted · say /unmute, ears lost · reconnecting, or push to talk · Enter opens the mic
  4. How long the last reply took to its first word, for 6 seconds after it starts
  5. The orchestrator's state: working, waiting on you or idle (see checking on a session)
──────────────────────────────────────────────────────▃ listening · 1.1 s to first word · Sequencer idle

While you type, the state line shows your line instead, until Enter:

▃ › /effort hi▏

Find past conversations

claudio saves every conversation as Markdown, one file a day:

~/.config/claudio/transcripts/2026-10-25.md

Each line has the time to the second and the same glyphs as the screen: your turns, claudio's sentences and what it showed, session reports, setting changes, notes, warnings, and each → command in full.

2026-10-25.md
14:02:11  › rebase 44 on main and run the tests
14:02:13  ◆ Sending that to Sequencer.
14:02:13    → cmux agent message 12BA24EB-6732-4999-BB16-7EE5AF2D35EE --from Voice "[spoken by the user, transcribed by claudio] rebase 44 on main and run the tests"
14:05:40  ▸ Sequencer  done  ·  for "rebase 44 on main and run the tests"
14:05:42  ◆ Rebased, tests pass, nothing to fix.

That send went to a cmux session. To a bb thread, the same line reads → bb thread tell thr_… "[spoken by the user, transcribed by claudio] …".

Ask "what did we do today?" or "what did I ask yesterday?" and the brain reads that day's file and answers from it, in order.

On this page