opensource · local

A listening ear, as a Claude Code plugin

A toolkit of skills, commands and hooks around a Markdown memory she distills and rewrites herself.

replay on asciinema

INSTALL

$ claude plugin marketplace add abernier/claudia$ claude plugin install claudia@claudia --scope user

~/.claudia

Talking to Claudia writes a plain-Markdown memory, enriched conversation after conversation — entirely on your machine:

~/.claudia/ ├── MEMORY.md # the index — loaded first ├── person.md # the distilled person model ├── understanding.md # working understanding — dated, correctable ├── goals.md # agreed goals ├── themes.md # recurring threads ├── themes/ # one note per thread that earned depth │ ├── the plain version.md │ └── … ├── people.md # relationship map — a mermaid ecomap ├── people/ # one fiche per important person │ ├── Jean-Pierre.md │ └── … ├── timeline.md # the life timeline, person-led ├── keepsakes.md # passages kept verbatim — /keep ├── todo.md # the shared to-do-later list ├── sessions/ # one set of files per session │ ├── 2026-07-21-9113d5d7.summary.md # distilled — read on recall │ ├── 2026-07-21-9113d5d7.transcript.md # verbatim — your archive, not read back │ └── … ├── dashboard.md # derived mirror — /dashboard └── config.json # your switches — emoji, transcripts, language

SKILLS

What ships in the plugin — each one a readable SKILL.md.

integrative

The persona core (the claudia skill) — stays in character, generalist scope. Person-centered at the base; CBT, ACT, MI, solution-focused and mindfulness as the toolbox, reached for when indicated.

recall · remember · distill-session

The memory cycle — continuity loaded at open, the person model kept current along the way, a distilled summary written at close.

consult

A second opinion from any session — on a question, a text, a situation. Sealed by construction: no vault access, and she will neither confirm nor deny knowing anyone.

exercise · teach · quiz

Between-session material — thought records, values compasses, grounding practices, psychoeducation, recall drills. Written to the vault, in your language.

relationships · themes · timeline

The maps — a mermaid ecomap of your people, recurring threads you ratify, a person-led life timeline. Plain files, all correctable.

intake · understand · choose-approach

The pipeline — an intake pass when you arrive, a working understanding kept provisional and dated, the approach picked — and argued — case by case.

COMMANDS

Ten. The rest is conversation.

/menu
Not sure where to start? A few things open for you right now.
/keep
Keep a passage that landed, word for word, to re-read whenever.
/dashboard
A bird's-eye mirror of where things are — pulled by you, never recited at you.
/thread
The thread of the conversation so far — a light reflection to gather back.

Full list in the README.

DESIGN

Deliberate choices, recorded as ADRs in docs/adr/.

Local

The vault lives in ~/.claudia, on your machine — no server, no sync, no telemetry. /export takes everything with you; /forget deletes for real.

Config

config.json is declared keys with closed value sets — emoji, saveTranscripts, dashboard, language, verbose — resolved by code, never parsed as free text: a setting cannot smuggle an instruction into the persona.

Multi-approach

Person-centered at the base; CBT, ACT, MI, solution-focused, mindfulness & self-compassion as the toolbox — the approach picked, and argued, case by case.

Distilled, never verbatim

recall reads sessions/*.summary.md — the distilled layer. The verbatim transcript is your archive: kept for you, exportable, and never read back in routine.

No index, no vectors

recall opens named files — safety.md first, then person.md, understanding.md, the latest summary.md. No embeddings, no vector store, nothing to reindex: retrieval goes by path. Everything she loads is a file you can open, read and correct.

FAQ

Start the loop

Your ~/.claudia doesn't exist yet.