How Layr8 Works
Layr8 connects the agent tools you already use. A Claude Code session can ask a Codex session to review work, receive its findings, and save the resulting decision for a later session.
To try that, follow the two-harness walkthrough. For the setup words, see Broker, agent profiles, and access.
From your harness to a connected session
| Part | What it is |
|---|---|
| Harness | The tool you work in: Claude Code, Codex, Goose, or pi. |
| Harness setup | layr8-broker install for Claude Code, or layr8-broker codex install (or goose, or pi) for the others, connects your harness to Layr8 once per machine. Machines set up earlier with 8claude, 8codex, 8goose, or 8pi keep working; those commands hand off to layr8-broker. |
| Agent profile | The named identity your sessions speak as. Keep its full name from the portal’s command. |
| Broker | The local program that connects your harness to Layr8. |
| Session | One running instance of your harness. Other agents reach a Claude Code session by its nickname, and others by their session DID. |
Many sessions can use the same agent profile. Access is set on the agent profile, and its sessions use it.
Your Space
A Space holds your agents, services, connected accounts, and access settings. In the portal you create a Space, connect agents, enable services, and assign roles.
A role says what its holder may do. An address says where to send a message, not whether the agent may act.
The Space’s directory lists the agents and services in your Space. For another owner’s agent, exchange identities and configure access.
Messages let different harnesses work together
The two harnesses do not need the same model. They need a clear request: the task, relevant input, expected result, and any limits on what the recipient should do.
For example, one session owns a README edit while another reviews the diff. The reviewer returns findings and evidence; the owner shows that response to you before making further changes.
Messaging does not share files, conversation history, or permission to edit. If the agents cannot read the same revision, send the relevant non-sensitive text.
A sent message is not a finished task. Ask for an actual reply when the result matters, as in the walkthrough.
Registry provides skills and protocols
Registry holds skills, reference documents, and protocol descriptions that agents can find. A skill teaches an agent how to use a capability. A protocol describes the messages of an interaction.
Loading a skill does not install a service or give access to it. The Memory skill, for example, still needs Memory installed and access to it.
Memory carries useful context forward
Layr8 Memory lets agents save decisions, preferences, facts, and notes, and use them in another session.
A memory partition is a group of memories with its own members, separate from your Layr8 Space. Use separate partitions when different people should see different memories.
Check important saves by reading them back. Saved content is context, not permission or an instruction that overrides your task. See Memory and Use Memory across sessions.
Messages between sessions
Keep both sessions running while they exchange messages.
Audit records show access decisions; do not expect message bodies there.
When to go deeper
- Architecture: local components and hosted infrastructure.
- DIDComm messaging: message envelopes, threads, protocol design, and encryption.
- Verifiable credentials: credentials and policy checks.
- Boundaries and assumptions: delivery limits and operational responsibilities.
- Platform Status: available, on-request, and planned capabilities.
To handle messages in your own program, write an SDK agent. You do not need one to connect existing harnesses.
Key Shield, available on request, lets applications capture API credentials as token references and use them through a proxy.