Hindsight

Use it when

• Your agent forgets user facts and prior decisions between sessions and you need durable per-user memory with metadata isolation
• Basic RAG over chat history retrieves irrelevant snippets and you need temporal, entity, and relationship aware recall plus reflection to form higher-level insights

What it solves

Not the fit when

• Treating `bank_id` or metadata filters as authentication or authorization; the application must derive tenant/user scope server-side and prevent callers from selecting another user's bank or filters
• Storing secrets or regulated personal data without defining retention, deletion, encryption, backup, and provider-processing policies
• Assuming repository LongMemEval results transfer to a different memory distribution or workload
• Simple stateless workflows where the operational and LLM-extraction overhead is unnecessary
• Assuming retain is free; it uses a configured local or hosted LLM provider
• vector search over static document corpora
• llm model serving
• chat ui hosting
• conversation transcript export

Install

docker run -p 8888:8888 -p 9999:9999 -e HINDSIGHT_API_LLM_API_KEY=$OPENAI_API_KEY -v hindsight-data:/home/hindsight/.pg0 ghcr.io/vectorize-io/hindsight:latest, then pip install hindsight-client (or npm install @vectorize-io/hindsight-client); embedded mode via pip install hindsight-all

Invoke

Python: client = Hindsight(base_url="http://localhost:8888"); client.retain(bank_id="my-bank", content="..."); client.recall(bank_id="my-bank", query="..."); client.reflect(bank_id="my-bank", query="...")

Alternatives

No reviewed alternatives recorded yet.