Familiars for Codex CLI: summonable terminal spirits with personality
What variant of Codex are you using?
CLI
What feature would you like to see?
An optional first-class Familiars system in Codex CLI: tiny discoverable terminal companions that live in the TUI, each with their own personality profile, small progression mechanics, and the ability to chime in when explicitly summoned.
The important part for me is discovery. I would not want familiars to be chosen from a picker at first launch. It feels much more fun if users stumble across them through normal usage: completed turns, tool use, debugging-heavy sessions, review sessions, long planning threads, and some token-driven “resonance.” When a familiar appears, you can bind it, keep it in a small grimoire/collection, and make one active.
Once active, a familiar could do a few things:
- show up as a small passive presence in the TUI footer or status line
- be summoned in-chat with
@name - support a small
/familiarcommand namespace likelist,status,pet,feed,dismiss, and maybesummon - stay low-noise by default, with commentary mostly explicit rather than constant
The vibe here is less “Tamagotchi glued onto a coding tool” and more “tiny techno-magical terminal spirits that feel native to Codex.”
A few example traits/stats could be things like:
- wit
- snark
- warmth
- debug-sense
- chaos
- focus
And different coding behaviors could attract different kinds of familiars. Someone doing lots of shell/debug work might encounter different familiars than someone spending most of their time reviewing diffs or planning architecture.
Why this feels like a first-class CLI feature rather than a plugin/skill:
- the TUI presence is part of the appeal
- summon syntax and structured mention flow would feel better if they were native
- the command and status-line integration seem like core CLI affordances
- if it works, it could still start in the CLI and later inform other Codex surfaces
I think this could be playful and useful: a familiar could add charm and personality to the CLI while also offering small context-sensitive reactions around debugging, review, or code smell moments without taking over the assistant’s voice.
Additional information
We’re prototyping this locally right now on our own branch/fork and are not assuming an external PR path, since the repo currently accepts unsolicited code contributions by invitation only.
If this direction is interesting, we’d be happy to share a more concrete implementation sketch for an MVP: discovery mechanics, data model, summon flow, TUI surfaces, and how we’d keep it fun without making the CLI noisy.
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗