Hydrate for GitHub Copilot · v1 launch

Your Copilot, minus the context tax.

Hydrate is a measured, auditable, on-premise memory and efficiency layer for AI assistants. Already shipping for Claude Code. The same binary - unchanged - extends to GitHub Copilot via a VS Code extension. Claude Code and Copilot share one memory when they both run on your machine: a decision captured from Copilot Chat injects into the next Claude Code prompt, and vice versa.

See the measured savings For Copilot Enterprise buyers

Ships with Hydrate v1 · on-prem by architecture · EU AI Act aligned

86-95%

token reduction across ten realistic Copilot scenarios, measured on two independent codebases. The 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.

Full methodology →

One engine, three surfaces

The same memory, wherever you work.

Hydrate doesn't care which assistant is asking. The on-prem binary answers context queries from any of three integrations - and every fact captured on one surface flows to the others.

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. Copilot chooses to call them when it needs prior context, wants to compress a long doc, or encounters a file its built-in read would struggle with.

Details ↓

MCP server

shipping

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

Integration guide →

Measured, not claimed

Two independent passes, same harness.

Both passes run the same ten scenarios (six recall queries, four compress calls on real project docs) and the same baseline assumptions. The only thing that changes between runs is the corpus - different codebases yield different numbers because the compress savings scale with input size, while recall savings stay flat.

Pass 1 · Hydrate repo 2026-04-21

~100 KB of project docs, short CLAUDE.md

Reduction @ M=3 88.0%

Reduction @ M=2 86.6%

Direct compress saving 10,903 tokens

Scenarios 10

Pass 2 · siteengine_ai repo 2026-04-22

34.5 KB CLAUDE.md plus 70+ KB of V5 spec docs

Reduction @ M=3 95.1%

Reduction @ M=2 94.9%

Direct compress saving 36,423 tokens

Scenarios 10

Reproduce on any codebase with scripts/measure-copilot.sh. Every figure is backed by a row in hydrate_retrievals; the id column in the measurement document links each number to a single HTTP request. Full methodology + per-row data.

How it works

Four surfaces Copilot already knows how to use.

Hydrate's VS Code extension registers a chat participant plus three Language Model Tools. Copilot decides when to call them - the same mechanism it uses for every other tool in its toolbox. No magic routing, no middleware, no new UI to learn.

@hydrate chat participant

Explicit invocation from Copilot Chat. @hydrate /recall <topic> returns exactly what Claude Code's hook would inject for the same query. /drain pulls any dashboard-queued thoughts. /capture flushes the session immediately.

→

`hydrate_recall`

Pulls prior decisions, session summaries, and pinned canon when the user references past work.

→

`hydrate_compress`

Compresses long prose through the local summariser before it reaches the model. 15×-100× ratios on real docs.

→

`hydrate_read_file`

Reads a workspace file, auto-compressing when it's bigger than 4 KB. Stops README.md blowing the context window.

The extension ships ~2,000 lines of TypeScript - a thin HTTP client over hydrate-server. Every Copilot-originated request carries X-Hydrate-Source: copilot; the server stamps the column server-side so savings attribute cleanly per editor. Source →

Copilot Enterprise buyers

Three walls every deployment hits between now and August 2026.

Token economics

The unit economics of the $39/seat Copilot Enterprise tier do not work at heavy-user volume without either a price rise or an efficiency lever. Hydrate is the efficiency lever - measured today, not projected. 86-95% reduction in context tokens per turn compounds across every seat in a large deployment.

EU AI Act - August 2026

Copilot Enterprise routes context through Azure OpenAI. Regulated customers in finance, healthcare and public sector are already refusing deployments where the context layer touches cloud. Hydrate's single on-prem binary, AES-GCM encryption at rest, and Article-mapped hydrate compliance report subcommand fill the gap without re-architecting Azure.

Customer-controllable compliance

Windows MSI, ADMX template for Group Policy, Intune/SCCM deployable. Storage mode, retention, forbidden paths, and API key all domain-configurable. Audit log rotates to JSONL for SIEM ingestion (Splunk, Sentinel, Elastic). Enterprise reviewers can reproduce every figure in the compliance report with one sqlite3 query.

Full enterprise page

Ships with Hydrate v1

Join the waitlist.

The Copilot extension launches alongside Hydrate's public v1 release. Waitlist opens Wednesday 6 May. Beta invites roll out in groups of 50 on a first-come-first-served basis; press, researchers and anyone writing about the launch can skip the queue via early access.

Wed6May

Waitlist opens Wednesday 6 May 2026.

Come back on the day to join - the beta rolls out in groups of 50 on a first-come-first-served basis.

Press, YouTuber, newsletter, podcaster, or researcher covering the launch? Request early access →