Feature request: per-repo instance isolation (data_dir, ports, and scoped export)

[maxlamagna]

Summary

This is a request for per-repo instance isolation — the ability to run agentchattr with a separate data_dir, ports, and scoped export for each project/repo.

Currently, the history store (data_dir) is shared across all repos. The repo-specific launcher/profile layer sets per-repo cwd for agents, but does not scope server.data_dir, ports, or the history/export store. This is a storage and isolation boundary problem, not just an export-format issue. The consequence is that /api/export returns all history from all projects, making it unsafe for per-repo disaster recovery or committed portable state.

Current behavior

• agentchattr-start-server --repo <repo> rewrites agent cwd via profile, but the server's data_dir (typically ~/ai-dotfiles/tools/agentchattr/data) is shared
• Ports are shared across all repos using the same agentchattr instance
• /api/export exports all messages, jobs, rules, and summaries across all repos/projects
• Exported message and job records do not carry repo or workspace metadata
• Channels help organize discussion within one project, but are not a durable repo partition across multiple projects

Desired behavior

Each repo/project should be able to run its own isolated agentchattr instance with:

• Its own data_dir (separate history store)
• Its own ports (no collisions between repos)
• Scoped export that includes only that repo's conversations

This would make per-repo DR safe (committing exported history to the repo) and prevent cross-project history leakage.

Suggested approaches

We identified three possible directions — any of these would solve the problem:

1. Per-repo instance isolation (recommended) — allow the launcher/profile layer to set a repo-specific data_dir and ports, so each repo runs its own server with its own store. This is the most practical option under user control today.

2. Repo-scoped storage/export in the server — add a workspace/repo field to the data model so the server can partition and export by repo natively. This would be the cleanest long-term solution.

3. Export filtering — add a query parameter to /api/export (e.g., ?channel=workflow,app-ui) to scope the export to specific channels. Less robust than options 1-2 but simpler to implement.

Context

We're using agentchattr with multiple AI agents (Claude, Codex, Gemini) in a Flutter app project. We recently built a committed portable DR layer for our issue tracker (Beads) and wanted to extend the same protection to our agentchattr chat history. The shared-store limitation is what blocked us from including agentchattr in that DR layer.

Acceptance criteria

• Each repo can run with its own isolated data_dir and ports
• Exporting from one repo must not include another repo's history
• A clear, documented path for per-repo persistence that works with the existing launcher/profile model

Thank you for building such a useful tool — this is a feature request to help us use it even more confidently across multiple projects!