Episodic memory · for Claude Code

Your AI forgets.
Hydrate remembers.

Claude Code starts each session with a blank mind. Hydrate captures what was built, decided, and named — then quietly injects it into the next session. No prompt engineering. No screenshots of previous chats. Just context that works.

$ curl -fsSL gethydrate.dev/install | sh
macOS · Linux · Windows
Get started — free Read the benchmarks
Works with Claude Code · data stays on your machine · no account required for Free
2 → 7
sessions shipping
Haiku, multi-session scenario
−57%
cost reduction
Opus, simple scenario
45%
of Sonnet cost
hybrid mode, Sonnet-tier quality
$0
forever
Free tier, local only

From the 2026-04-17 benchmark run. $67 of real API spend. Every number is reproducible.

What changes

The difference is execution, not intelligence.

Models don't get smarter with Hydrate. They get context. That's usually enough.

without Hydrate 
$ claude "add a dark mode toggle"

Haiku: What should the default
 colour scheme be?
Haiku: What do you mean by
 "toggle" — a button? switch?
Haiku: Which pages should be
 affected?

[ nothing written ]
with Hydrate 
$ claude "add a dark mode toggle"

Haiku: edits src/styles/globals.css
Haiku: edits src/components/Nav.tsx
Haiku: Done. Toggle wired to
 prefers-color-scheme. Persists in
 localStorage. Icon updates on
 change.
One intervention · four outcomes

It pays off differently for every model.

Hydrate isn't a uniform speed-up. Each model has a different failure mode without context. Hydrate removes all four — at different costs, with different value propositions.

Haiku 4.5 reliability
before
2/7 sessions ship. Asks clarifying questions instead of building.
with Hydrate
7/7 sessions ship. Executes every prompt.
Sonnet 4.6 −14% cost
before
Builds cleanly, 7/7 sessions.
with Hydrate
Builds cleanly, modest efficiency gain.
Opus 4.7 (simple) −57% cost
before
Re-explores workspace every session.
with Hydrate
Skips re-exploration, produces same output.
Opus 4.7 (complex) 2× features
before
3/7 sessions refuse with "need clarification".
with Hydrate
6/7 sessions build, converting refusals into features.
Hybrid — Sonnet s1 → Haiku s2+ ★ flagship
before
n/a — this is a workflow, not a model.
with Hydrate
Sonnet-tier design propagates to Haiku via memory. $1.41 total.
See the full benchmarks
How it works

Three quiet pieces.

01

Hooks

Hydrate installs two Claude Code hooks — UserPromptSubmit and Stop. They're invisible to you. They capture each session's prompt, decisions, and output.

02

Facts

Between sessions, an extractor pulls concrete facts — "pricing tiers are Free/Pro/Team", "accent colour is #89b4fa" — and stores them scoped to your project.

03

Dashboard

A local dashboard at localhost:8089/dashboard shows your spend, cache ratio, and facts per project. Nothing ever leaves your machine — Free, Pro, or Team.

Pricing

Start free. Pay only when you need the rest.

Free

$0/ forever

Up to 2 active projects, full hooks + dashboard. No account.

Install
Coming in ~3 weeks

Pro

$9/ month

Unlimited active projects plus export / import.

Join Pro waitlist →

Team

$29/ user / month

Shared project memory across teammates. Hybrid mode shines here.

Join Team waitlist →

Enterprise with SSO, self-host, org-scoped facts — contact sales.

Ready when you are

Install in 30 seconds.

No account. No telemetry. The Free tier never touches the network.

$ curl -fsSL gethydrate.dev/install | sh