One brain. Every runtime keeps its own model bill.
The same memory across Claude Code, OpenAI Codex (CLI and app) and Mistral Vibe today, with Hermes in alpha. Each runtime authenticates and pays for its own model; Hydrate is the local-first layer they share. Pin a fact in one and a fresh session in another recalls it, automatically in Claude Code and Codex and on demand everywhere else, without sending the conversation to any cloud Hydrate operates.
The wedge: every competitor is a single-runtime silo.
No other tool in the audited 2026-Q2 landscape demonstrates capture in one runtime and recall in another on a single machine without funnelling the conversation through a third-party hosted memory service. Because Hydrate is a layer rather than a runtime, an Author in one model family and a Critic in another can work over the same shared state, which is exactly what makes cross-model orchestration possible.
Runtimes today
Claude Code
full auto-memoryNative hooks: inject, capture and MCP all reach the model. Zero configuration, works on day one with your existing Claude subscription.
Codex CLI + app
full auto-memoryOpenAI's Codex CLI and macOS app share the same local store, with inject, capture and MCP all reaching the model. Authenticates via your ChatGPT subscription, not Hydrate.
Mistral Vibe
full (fork)Wired via hydrate vibe install. Receives shared project memory on the first tool call and contributes facts back. Automatic session-start injection needs the Vibe fork.
Hermes
alphaNous Research's Hermes via an exclusive memory-provider plugin slot (early access). Reads shared identity into the system prompt; reads and writes project recall against the local daemon.
Alpha caveats: profile write-back is inert (v0.2), identity facts only prime other runtimes when pinned, and gateway binding needs broader live testing.
VS Code + Copilot
capture + MCPExtension with an @hydrate chat participant plus Language Model Tools Copilot auto-invokes. Capture plus on-demand recall; no automatic session-start injection.
MCP clients
capture + MCPEvery MCP-aware client: Claude Desktop, Cursor, Antigravity, Cline, Zed, Gemini CLI, your own agent. One endpoint, the same memory, over on-demand recall.
The tiers are real, and we mark our own gaps. Claude Code and OpenAI Codex are the verified full auto-memory story: inject, capture and MCP all reach the model every session. Mistral Vibe is full through its fork. Copilot and the other MCP clients capture every session and serve recall on demand, but a host that withholds injected context means no automatic session-start recall there yet. Per-runtime evidence is in RUNTIME_INTEGRATION.md.