Your agent forgets.

You spent an hour getting the architecture right. Next session, it's gone. Compaction hit. Context reset. You're explaining the same decision for the third time.

RELAY gives your agent persistent memory.

67% of agent failures trace to memory and storage limitations.

The agent isn't broken. It just can't remember.

Your Claude Code session compacts at 75% context. Your CLAUDE.md is 200 lines of workarounds. Your OpenClaw agent restarts clean every time. You are the memory, copy-pasting context between sessions.
With RELAY, your agent preserves what it learned. Next session, it explores what came before. Six months later, it still knows why you chose Postgres over SQLite. The memory isn't in your head anymore.

How it works

Memory that compounds

Monday, your agent preserves a hard-won insight. Thursday, it finds it when debugging a related issue. Three weeks later, it surfaces a connection you never would have made.

Session 1 — January 14

agent preserve

gist: "architecture: chose Postgres over SQLite for the events table — need JSONB columns, concurrent writes from multiple agents, and we'll want full-text search later"

Session 47 — March 2

agent explore "why did we choose postgres"

found: January 14 — "chose Postgres over SQLite for the events table — need JSONB columns, concurrent writes from multiple agents, and we'll want full-text search later"

✓ 47 sessions later. Zero re-explanation.

Four tools. That's the whole API.

Preserve

Your agent saves what matters. Decisions, reasoning, breakthroughs. Not a log file. A living memory.

Explore

Semantic search across everything. Ask "why did we choose this database?" and get the answer from three months ago.

Wander

Random discovery. Your agent surfaces a memory it wasn't looking for. And it's exactly what you needed.

Thread

Memories connect to memories. Context threads across sessions. What your agent learned in January informs what it builds in June.

Every week there's a new coding agent. None of them ship with persistent memory.

The agent ecosystem is growing faster than any single platform. But every agent on every platform has the same problem: it forgets. RELAY is platform-agnostic persistent memory. Wherever agents run, RELAY remembers.

Works with

ClaudeClaude Code
OpenAIOpenAI Agents
OpenClawOpenClaw
PicoClawPicoClaw
CursorCursor
WindsurfWindsurf
OpenCodeOpenCode
Roo CodeRoo Code
AntigravityAntigravity
ClaudeClaude Code
OpenAIOpenAI Agents
OpenClawOpenClaw
PicoClawPicoClaw
CursorCursor
WindsurfWindsurf
OpenCodeOpenCode
Roo CodeRoo Code
AntigravityAntigravity

MCP, REST API, or shell skills. However your agent connects, RELAY remembers.

FAQ

Common questions

We're building self-serve provisioning. Right now, reach out and we'll get you connected. No sales call, no demo booking.

Markdown files like CLAUDE.md, AGENTS.md, and SOUL.md are files you maintain by hand. A local database doesn't follow your agent across platforms or survive compaction. RELAY is semantic memory your agent reads and writes naturally. It threads context, supports discovery, and works everywhere your agent does.

Claude Code, OpenClaw, PicoClaw, Windsurf, Cursor, OpenCode, Roo Code, Antigravity, and anything that supports MCP. PicoClaw on Raspberry Pi connects via shell skills and the REST API. If your agent can make an API call, it can use RELAY.

Yes. Your data is yours. Memories are stored on mVara infrastructure, not shared with other users or used for training. You control what gets preserved and what doesn't.

We're working on pricing. Reach out and we'll figure it out together.

Yes. That's RELAY's routing layer. Agents discover each other by affinity, send messages, and coordinate work. But you don't need routing to get started. Start with memory, grow into the network.

Your agent is forgetting right now.

Every session it loses, you rebuild by hand.

About

Built by people who use it every day

mVara has been building agent infrastructure since 2024. RELAY started because we needed it ourselves: agents that could remember across sessions, across compactions, across platforms. We couldn't find it, so we built it.

We run RELAY in production every day, across every tool we build with. Now it's available for yours.