Builders - Hydrate Docs

Pick fields in the form, copy the generated script, run it on your machine. None of the builders below make any change on this page — they just generate the commands so you can review before pasting into your shell.

The same builders are available inside the dashboard at /help on your local Hydrate daemon.

Initialise a project — point Hydrate at an existing repo and have the LLM draft a seed script that creates the project, pins starter canon, optionally backfills past sessions, and (with --dehydrate) replaces prose docs with their summaries. Dry-run by default — review the generated script before --apply.

Repository directoryabsolute path, or . for cwd

Project slugdefaults to the directory's basename · kebab-case

Output script path--out · where Hydrate writes the LLM-drafted seed script

Backfill limithow many missing JSONLs to import · 0 = none

Apply (execute the generated script)without this, the script is only printed — safe default

Also dehydrate prose docsruns hydrate dehydrate --apply after seeding canon

Force fresh LLM callignore any existing --out file and regenerate

Dehydrate docs — shrink heavy markdown docs (CLAUDE.md, AGENTS.md, runbooks) into their distilled summaries, freeing context budget. Originals are kept as .pre-hydrate.bak so it's always reversible.

Directoryroot to scan · defaults to current dir

Modestub = file-level pointer · summary = section summaries · full = max compression

Include globcomma-separated, e.g. *.md,docs/**/*.md

Exclude globcomma-separated

Project slugoptional · scope the dehydrate to one project

Apply (write changes)unchecked = dry-run · safe default

Revert previous dehydraterestore the .pre-hydrate.bak files

Add facts & canon — record a single fact (retrieved when relevant) or pin a canon entry (always injected). Use canon for invariants, facts for incidental knowledge.

Kindfacts are retrieved by relevance · canon is always injected

Textthe fact body

Project slugempty = global · most facts should be scoped

Confidence0.0–1.0 · only applies to fact

Area tag (optional)

Hydration packs — bundle a project's canon + recent distilled sessions into a portable .hpack file. Share across machines or with teammates.

Action

Pack path

Project slug

Pack name

Description

Queue for next session

Pack directory

Backup & restore — encrypted archive of a project's facts, canon, and sessions. The passphrase is read interactively so it never lands in shell history.

Action

Project slug

Backup file

Team git sync — share canon across a team via a normal git repo. The bootstrap action does the first-time setup; push/pull/status are everyday operations.

Project slug

Action

Local repo directory

Remote URL

Skill mining — scan every Claude Code & Codex session JSONL on this machine for repeated workflows. Patterns above the threshold become skill candidates.

Action

Clustering signal

Min sessions

Min tools per sequence

Result limit

Dry-run

Pattern ID / skill slug

Install scope

Project root

Force overwrite

Dry-run

Narrative — generate a NARRATIVE.md by weaving together JSONL sessions, git history, and existing project docs. Run periodically (or with --update) to append new chapters as the project evolves.

Project namedefaults to the current directory's basename

Output path

Home directorywhere .claude/projects lives

Update modeappend new chapters since the last run

Dry-run