跳到主要内容

Use with self-hosted Hindsight

If you run the open-source Hindsight engine yourself, you don't have to host embedding, rerank or decision models to get good retrieval. Point its providers at this API and it uses the same models Hindsight Cloud runs — with one API key and one shared credit balance, and no GPUs of your own.

Hindsight 1.2 and later: sign in once​

Hindsight 1.2 has a hindsight provider for all three models. Sign in once and it picks up the key, the base URL and the model names by itself:

hindsight-admin cloud login     # signs in to Hindsight Cloud and saves a key
export HINDSIGHT_API_DECISION_MODEL_PROVIDER=hindsight
export HINDSIGHT_API_RECALL_DECISION_MODEL_PROVIDER=hindsight   # optional: rank recall with it too
export HINDSIGHT_API_RERANKER_PROVIDER=hindsight                 # optional
export HINDSIGHT_API_EMBEDDINGS_PROVIDER=hindsight               # optional

The decision model then runs consolidation, reflect and knowledge pages. Instead of signing in, you can set HINDSIGHT_API_DECISION_MODEL_API_KEY with an inference-capable API key (see Authentication). For every setting, see Decision Models.

Any version: wire each provider​

Each model is wired independently, so enable any subset. You need an inference-capable API key (see Authentication) and the base URL:

https://inference.hindsight.vectorize.io

Embeddings​

The OpenAI embeddings provider, pointed at /v1/embeddings:

HINDSIGHT_API_EMBEDDINGS_PROVIDER=openai
HINDSIGHT_API_EMBEDDINGS_OPENAI_BASE_URL=https://inference.hindsight.vectorize.io/v1
HINDSIGHT_API_EMBEDDINGS_OPENAI_MODEL=hs-embed-1
HINDSIGHT_API_EMBEDDINGS_OPENAI_API_KEY=<your Hindsight API key>
HINDSIGHT_API_EMBEDDINGS_OPENAI_BATCH_SIZE=128

Reranking​

The Cohere reranker provider. Unlike the Cohere SDK, Hindsight's provider uses the base URL verbatim, so include /v1/rerank:

HINDSIGHT_API_RERANKER_PROVIDER=cohere
HINDSIGHT_API_RERANKER_COHERE_BASE_URL=https://inference.hindsight.vectorize.io/v1/rerank
HINDSIGHT_API_RERANKER_COHERE_MODEL=hs-decide-1-flash
HINDSIGHT_API_RERANKER_COHERE_API_KEY=<your Hindsight API key>

Decision model​

The TypeSafe decision provider — used for consolidation, reflect and mental-model refresh. Give it the bare host (no /v1); the provider appends the System One path itself:

HINDSIGHT_API_DECISION_MODEL_PROVIDER=typesafe
HINDSIGHT_API_DECISION_MODEL_BASE_URL=https://inference.hindsight.vectorize.io
HINDSIGHT_API_DECISION_MODEL_MODEL=hs-decide-1-flash
HINDSIGHT_API_DECISION_MODEL_API_KEY=<your Hindsight API key>

Every call still bills against your shared balance at the standard rates, and GET /v1/models lists exactly what your key can reach. The model IDs above are the current defaults; see Models for the live list.