BETA In open beta. Install live. Lock $5/mo for your first 12 months. See pricing →
Hydrate for Claude Code · shipping now

Claude Code, with persistent memory.

Hydrate is a measured, auditable, on-premise memory layer for Claude Code. Four native hooks wire into the Claude Code lifecycle to inject prior context before every prompt and capture decisions after every session. No configuration. No cloud. Works on day one with your existing Claude subscription.

Shipping now · on-prem by architecture · EU AI Act aligned
86-95%
token reduction across ten realistic scenarios, measured on two independent codebases. The Hydrate compress path alone saves 10,903 - 36,423 tokens per ten-scenario run as a direct measurement: input and output both observed, no baseline assumption.
How it works

Four hooks, zero configuration.

Hydrate installs four Claude Code hooks. Each fires at a defined point in the Claude Code lifecycle. Together they form a complete memory loop: inject before the prompt, capture after the session, enrich before tools, observe after tools.

UserPromptSubmit Before every prompt

Injects relevant prior decisions, session summaries, and pinned canon into the prompt context. Zero tokens wasted on things Hydrate already knows.

Stop After session ends

Captures the finished session: extracts facts, decisions, and open threads, then writes them to the local store. Nothing lost between sessions.

PreToolUse Before every tool call

Optionally enriches tool context with file-level or task-level memory before Claude Code touches the filesystem. Keeps large-project context sharp.

PostToolUse After every tool call

Observes tool results to update in-session working memory. Useful for tracking which files changed and what decisions were made along the way.

One binary, one config file

Running hydrate setup drops four entries into your ~/.claude/settings.json and starts the local server on localhost:24312. That is the entire setup. No daemon process manager, no Docker, no cloud account required.

The hooks communicate with hydrate-server over localhost HTTP. Session data stays on your machine: SQLite at ~/.hydrate/sessions.db, encrypted at rest with AES-GCM. Nothing leaves the machine unless you are on Pro tier with a BYOK API key and have opted in to fact extraction.

Install guide →
One engine, three surfaces

The same memory, wherever you work.

Hydrate does not care which assistant is asking. The on-prem binary answers context queries from Claude Code, VS Code Copilot, and any MCP client. Every fact captured on one surface flows to the others.

Three CLI agent windows side by side (Claude Code by Anthropic, Codex CLI by OpenAI, and Mistral Vibe by Mistral AI). Each shows its session starting (analyzing repository, scanning workspace, indexing project) followed by the same Hydrate context block being injected (User prefers Go with explicit error handling; Project uses zap.Sugar for structured logging; Do not touch files in /legacy without asking; Tests live in /internal/*_test.go). Each card shows a 'hydrate ↓ 4 facts' badge in the bottom-right. All three agents share the same central memory cylinder labelled ~/.hydrate/data.db.

Claude Code

shipping

Native hooks - UserPromptSubmit, Stop, PreToolUse, PostToolUse. Zero configuration. Works on day one with your existing Claude subscription.

Install guide →

VS Code + Copilot

v1 launch

VS Code extension registering an @hydrate chat participant plus three Language Model Tools. Same memory store as Claude Code - decisions captured in one flow to the other.

Copilot page →

MCP server

shipping

Works with every MCP-aware client: Claude Desktop, Cursor, Cline, Zed, Gemini CLI, Mistral Vibe, your own agent. One endpoint, the same memory.

Integration guide →
Local-first by architecture

Your sessions never leave your machine.

01

On-prem binary

hydrate-server runs on your machine. There is no cloud endpoint for session data. The free tier never makes an outbound network call beyond licence validation.

02

Encrypted at rest

The local SQLite store is encrypted with AES-GCM. Keys are derived from your machine identity. Even if the file is copied, it cannot be read on another machine without the same key material.

03

Secrets scrubbed

The internal/scrubber package strips common secret patterns - API keys, tokens, private key blocks - before any storage or LLM processing. Credentials typed in a session are never persisted or transmitted.

04

EU AI Act aligned

Hydrate ships an Article-mapped hydrate compliance report subcommand. Every figure in the report is backed by a row in hydrate_retrievals and can be reproduced with a single sqlite3 query.

Available now

Install Hydrate today.

Hydrate is available for Claude Code on macOS and Linux. Free tier is unlimited local memory with no account required. Pro tier features (BYOK fact extraction, Team dashboard) ship alongside the public v1 launch.