Running Pluribus: use Docker Compose from the repo (see root README.md) for a local API, or point pluribus.baseUrl at another host via user/local settings. Do not commit site-specific IPs or LAN addresses into the repository.
The in-repo Pluribus AI extension (identifier pluribus.pluribus-ai; packaged as pluribus-ai-<version>.vsix) under integrations/vscode/extension/ is an event-driven orchestrator:
- Health: periodic
GET /healthz, status bar + sidebar. - Auto recall:
POST /v1/recall/compileon task start and debug start (optional debounced save). - Auto record:
POST /v1/advisory-episodeson task process non-zero exit (source: vscode-orchestrator); optional debug end (default off). - Metrics: local counters + Output logging — no network telemetry.
- Policy: no client-side memory ranking or promotion; all semantics remain on the control plane.
Details, settings, and .vsix packaging: integrations/vscode/README.md.
VS Code’s MCP story is vendor-specific. Use integrations/vscode/mcp-config.example.json and your Copilot/MCP extension’s docs. Endpoint: http://127.0.0.1:8123/v1/mcp (or deployed host).
Copy integrations/vscode/github-copilot-instructions.template.md → .github/copilot-instructions.md; merge snippets/context-prime.txt. Skill table: integrations/vscode/skill.md.
The same .vsix may install in Cursor; behavior should be verified manually (tasks/debug events). VS Code is the tested reference for this repository.
- Auto record on task failure requires
onDidEndTaskProcess(tasks with a real process). - Save-triggered recall is off by default (noise).
- Hard enforcement is intentionally out of scope — soft nudges and defaults only.