Initial public release
Hydra is a context hydration engine for second brains. It scans the artifacts agent-assisted work leaves behind — plans, session recaps, decision records, changelogs, git history, loose files — classifies and enriches them, then projects them into an Obsidian vault that both humans and agents can query.
Highlights
- End-to-end hydration pipeline in
HydraCore: scan → classify (ArtifactKind) → enrich (tags, links, provenance, authority) → SQLite outbox → deterministic PARA-style vault projection. - Three surfaces over one graph: a SwiftUI + TCA app, the
hydraCLI (scan,health,hydrate,search,serve), and a JSON-RPC 2.0 MCP server on Hummingbird for Claude's stdio transport. - Dry-run first:
--dry-runand the in-app dry-run preview let you inspect projected notes before anything is written to a vault. - Snapshot-tested UI: 40 test functions across four targets, including committed light/dark reference images for every delivery state, confidence level, and hydration stage. Every view ships with an Xcode
#Preview. - New portfolio README with an animated banner, a pipeline architecture diagram, and an MCP sequence diagram.
Getting started
Requires macOS 15+ and a Swift 6.2 toolchain.
git clone https://github.com/Ripnrip/Hydra.git
cd Hydra
swift build && swift test
hydra health --vault ~/Documents/MyVault
hydra hydrate --source ~/.claude/plans --vault ~/Documents/MyVault --dry-runSee the README for the full walkthrough.
Notes
This is an early foundation, not a finished product, and the code is explicit about the gap:
- 2 of 8 MCP tools are implemented.
health_checkandsearchwork;hydrate,relationships,gaps,timeline,tag_reportandprojectNoteare declared and returnnot_yet_implemented. - 2 of 8 delivery states are mechanically enforced. Only
submittedandblockedare tracked; the other six are policy-defined and flagged viaisMechanicallyTracked. - macOS 15+ / Swift 6.2 only. No prebuilt binary is attached — build from source.
- Use
--dry-runbefore pointing Hydra at a vault you cannot afford to lose.
Developed against a live vault of 331 notes, 685 tags, PARA structure, and 16 Obsidian plugins.