Hindsight
MemoryMIT
Hindsight is an agent memory system from Vectorize for agents that learn over time rather than only recalling past conversations. Its core operations are retain, recall and reflect. Developers run the server with Docker, pip or Helm, store data in PostgreSQL with pgvector, and connect through a REST API, an MCP endpoint per memory bank, Python, TypeScript or Go clients, or a command line tool. A managed Hindsight Cloud service is also offered, and an LLM provider key is required.
The server exposes an API at localhost:8888 and a UI at localhost:9999 in the quick start, with one MCP endpoint per memory bank.

Pinout
- MCP server: supported, verified
- HTTP API: supported, verified
- command line: supported, verified
- Python SDK: supported, verified
- TypeScript SDK: supported, verified
- OpenAPI description: not checked yet
| Interface | Status | Source we read | Read on |
|---|---|---|---|
| MCP server | Supported, verified | GitHub README, vectorize-io/hindsight, Interfaces | |
| HTTP API | Supported, verified | GitHub README, vectorize-io/hindsight, Interfaces | |
| Command line | Supported, verified | GitHub README, vectorize-io/hindsight, Interfaces | |
| Python SDK | Supported, verified | GitHub README, vectorize-io/hindsight, Interfaces | |
| TypeScript SDK | Supported, verified | GitHub README, vectorize-io/hindsight, Interfaces | |
| OpenAPI description | Not checked yet | — | — |
Facts
| Pricing | Hindsight Cloud is pay as you go with no monthly fee and $5 of free credit: retain $10 and recall $0.75 per million tokens, reflect $0.05 per call. Vectorize pricing page, read . |
|---|---|
| Purpose | An agent memory system for agents that learn over time. GitHub README, vectorize-io/hindsight, read . |
| Operations | retain (store), recall (search) and reflect (deeper analysis over existing memories). GitHub README, vectorize-io/hindsight, read . |
| Server install | Docker image ghcr.io/vectorize-io/hindsight, pip install hindsight-api, or a Helm chart. GitHub README, vectorize-io/hindsight, Install, read . |
| Clients | pip install hindsight-client, npm install @vectorize-io/hindsight-client, a Go client and a CLI. GitHub README, vectorize-io/hindsight, Interfaces, read . |
| REST API | API at http://localhost:8888 and UI at http://localhost:9999 in the quick start. GitHub README, vectorize-io/hindsight, Interfaces, read . |
| MCP | One MCP server endpoint per bank, enabled by default, at /mcp/{bank_id}/. GitHub README, vectorize-io/hindsight, Interfaces, read . |
| Storage | PostgreSQL with pgvector, or Oracle AI Database 23ai; an embedded database (pg0) is listed. GitHub README, vectorize-io/hindsight, Dependencies, read . |
| Hosted option | Hindsight Cloud is a managed, usage-based service. GitHub README, vectorize-io/hindsight, Hosted vs self-hosted, read . |
| License | MIT. Repository LICENSE file, read . |
| Last commit | . GitHub API, latest commit on the default branch of vectorize-io/hindsight, read . |
Limits
- An LLM provider API key is required to run the server.
- Decision, rerank and embedding models are published separately as Hindsight Models.
Strengths
- Python, TypeScript, Go, REST, MCP and command line interfaces are all documented.
- Both self-hosted and managed options exist.
- The repository is under the MIT license.
Pricing
As written on Hindsight's own page on the read date. Prices change: check there before you pay.
Alternatives to Hindsight
All memory tools for agents
Open-source memory layer storing agent context across sessions and models.

Memory layer that wraps LLM clients to store and recall agent state.

Memory API with a hosted MCP server, Python and TypeScript SDKs, and a local server.

Open-source memory platform that turns documents and conversations into searchable memory.
Last check: . Method: public repository and docs read automatically, page and day written down. Next check due within 90 days. How we check
Is this your tool? Write to hello@agent-tools.directory from an address on its domain to correct a pin or a fact.