> ## Documentation Index
> Fetch the complete documentation index at: https://docs.conare.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Core concepts

> Integrations, end-user IDs, per-source containers, opaque tenancy, and the three retrieval modes — search, deep recall, and suggestions.

## Integration

The durable boundary between your product and Conare. An Integration is owned by your organization; separate deployment stages (staging, prod) are separate Integrations disambiguated by slug. Each has its own `cint_...` key with explicit scopes (`memory:read`, `memory:write`, `memory:delete`).

## End users

Every memory call names an `endUserId` (1–128 chars, `A-Za-z0-9@._-`) — your identifier for your user. It is **not** a namespace: the physical tenant is an opaque HMAC derived from your Integration and the end-user ID. Clients cannot select namespaces or address another tenant, by construction.

* Set display identity with `PUT /api/v1/users/{endUserId}` (name/email for your dashboard views).
* `DELETE /api/v1/users/{endUserId}` is a full GDPR wipe of that user.

## Containers

Containers group one user's memories by source — e.g. `profile`, `claude-chats`, `saved`, or one per connected data source (container tag = connector id). Use `containerTag` on save to organize, and `DELETE /api/v1/containers/{containerTag}` to drop one source's memories without touching the rest.

## Retrieval modes

| Mode        | Endpoint                   | Latency    | What you get                                                                           |
| ----------- | -------------------------- | ---------- | -------------------------------------------------------------------------------------- |
| Search      | `POST /api/v1/search`      | sub-second | Ranked raw matches (vector + BM25, RRF-fused, reranked). No LLM.                       |
| Deep recall | `POST /api/v1/recall`      | \~3–6s     | Synthesized, citation-grounded answer as prompt-injectable text.                       |
| Suggestions | `POST /api/v1/suggestions` | \~3–6s     | Up to 5 concrete next actions, each grounded in a specific memory. `[]` for new users. |

## Errors and observability

Errors use a stable envelope — `{ statusCode, code, message, requestId, details? }` — and never leak internal tenant IDs or backend text. Every response includes `X-Request-Id`; send your own to correlate end-to-end. Rate limits are signaled with standard headers and `429`; usage-ledger plans signal an exhausted allowance with `402` plus balance/reset metadata.