Claude Code self-hosted environments decide where a coding session runs. mxLore decides where the specs, decisions and lessons from those sessions are kept. Different layers, no overlap, and they work together.
Independent project. mxLore is not affiliated with, sponsored by, or endorsed by Anthropic. Claude and Claude Code are trademarks of Anthropic, PBC. Statements about Claude Code below are taken from Anthropic's own announcement and documentation, linked in the sources, and reflect the public beta as of September 4, 2026.
A Claude Code session has three parts that can live in three different places. Self-hosting moves the middle one. mxLore is the bottom one, and it stays put no matter where the session runs.
Where prompts, responses and tool results are processed.
Where the checkout, build artifacts and files a session creates are kept.
Specs, decisions, lessons, the graph between them, and messages between sessions.
Read it top to bottom. Inference is Anthropic's in every column. The runtime column you pick is a question of network, compliance and plan. The knowledge row does not change with that choice, and that is the whole point: a session that starts in the cloud on Monday and on a self-hosted runner on Tuesday reads the same decisions and writes to the same graph.
Each Claude Code cell below is Anthropic's own description of the feature, with a numbered source. Each mxLore cell describes what the server does.
| Item | Claude Code cloud session | Claude Code self-hosted environment (beta) | mxLore |
|---|---|---|---|
| Model inference | Processed by Anthropic1 | Processed by Anthropic1 | Does not run a model. Your coding AI reads mxLore through MCP tools and reasons with whatever model it uses. |
| Repository checkout, build artifacts, secrets, files a session creates | Anthropic-managed environment1 | Infrastructure you provision1 | Outside mxLore's scope. It does not touch your checkout. |
| Conversation transcript | Stored by Anthropic so a session can be resumed from any surface1 | Stored by Anthropic so a session can be resumed from any surface1 | Stored on your server only if your AI writes a session note. Transcripts themselves are not collected. |
| Specs, decisions, plans, lessons that outlive a session | Outside the scope of this feature | Outside the scope of this feature | Structured documents in your MariaDB, with revisions, relations and a knowledge graph, shared across projects and developers. |
| Recall of past lessons when a file is touched | Outside the scope of this feature | Outside the scope of this feature | Recall gate: lessons linked to the file are surfaced before the edit. |
| Messages between sessions | Cross-session messaging in Claude Code 2.1.224: sessions on the same machine discover each other and exchange text. macOS and Linux.2 | Messages are stored on your server and delivered to the receiving session by its local proxy, so the two sessions can be on different machines. Messages wait until acknowledged and can point to a document by id. | |
| Who can turn it on | Included in Claude Code cloud sessions2 | Team and Enterprise plans, public beta. Off by default, an Owner enables it in admin settings. Not available to organizations using zero data retention.1 | Any plan, any MCP client. You install the server, the proxy and the skills. |
| Setup | Start a session from claude.ai, the apps or claude --cloud2 |
Run claude self-hosted-runner on your machines or containers; fixed or on-demand runners1 |
One Windows executable plus MariaDB, a stdio proxy for Windows and macOS, and a skill bundle for Claude Code. |
The release that introduced self-hosted environments also introduced messaging between Claude Code sessions. It overlaps with something mxLore has done for a while, so here is the difference, plainly.
Claude discovers your other sessions with a list-agents tool and sends text with a send-message tool, on request or on its own. The message is text Claude writes for the other session, never the conversation history or files.
Available on macOS and Linux, between sessions on the same machine, from version 2.1.224.
A session stores the message on your server. The local proxy of the receiving session polls the server and writes it into a per-project buffer, and a background watch wakes the session the moment it changes. The two sessions can sit on different machines.
Messages stay on the server until acknowledged, carry an optional document id instead of a pasted body, and are scoped to a project so a machine shared by several projects never mixes them up.
Three situations where the runtime choice and the knowledge layer solve different problems at the same time.
A self-hosted runner puts the session next to your internal services. mxLore, also on your network, gives that session the decisions and lessons your team recorded before it started.
One developer works in a cloud session, another in a terminal, a third on a runner. All three read and write the same specs, and a lesson learned in one place is recalled in the others.
Transcripts let you resume a session. mxLore lets the session after that, and the one a month later, start from what was decided rather than from what was said.
Last reviewed: September 4, 2026. Self-hosted environments are a public beta and Anthropic may change plans, platforms or behavior. If a statement on this page no longer matches the linked sources, the sources win. Tell us at info@microtronx.com and we will correct it. mxLore statements describe the server as documented at github.com/MicrotronX/mxLore.