← ClaudeAtlas

aegis-setuplisted

Set up AegisDB as persistent memory for this Claude Code project. Asks a few short questions (local vs shared server, embeddings, capture quality, auth), then scaffolds .mcp.json and the recall/capture hooks via `aegisdb-init`.
d4n-larsson/aegisdb · ★ 6 · AI & Automation · score 74
Install: claude install-skill d4n-larsson/aegisdb
# Set up AegisDB memory for this project Your job is to wire AegisDB in as this project's persistent memory by gathering a few settings and then running the `aegisdb-init` scaffolder. Do **not** hand-edit `.mcp.json` or `.claude/settings.json` yourself — `aegisdb-init` writes them correctly and idempotently. Keep the conversation short. ## 1. Gather the settings Ask the user these questions (offer the defaults; accept "just use defaults"). If they passed arguments, treat `$0` as the host and `$1` as the port. 1. **Where is the server?** - *Local, and I need one* → you'll start one in step 2. - *Local, already running* → host `127.0.0.1`, port `9470`. - *Shared / team server* → ask for the host and port, and the **auth token** they were given. (With a namespaced token they do **not** need a namespace.) 2. **Embeddings** (controls semantic recall; the dimension MUST match the server's `--embedding-dim`): - `none` (default) — tag/time recall only, no setup. - `voyage` — best recall; dimension `1024`; needs `VOYAGE_API_KEY` in the environment. - `local` — offline; dimension `384`; downloads a model on first use. 3. **Capture quality** (how finished sessions become memories): - `none` (default) — heuristic: keep salient sentences from the transcript. - `claude-code` — distil the session into durable facts with an LLM (dedup + supersede contradictions on write), reusing the local `claude` CLI auth — **no API key needed**. Best qual