Hindsight

Hindsight is the default memory provider: Docker Compose runs LiteLLM, the Hindsight API, and the shared indexer on loopback. Retain, recall, reflect, and consolidate go through Hindsight’s HTTP API; LiteLLM routes model calls to lemond, Ollama, or another OpenAI-compatible backend.

Upstream project: vectorize-io/hindsight.

What personal-memory runs

Service	Role
`litellm`	Proxy for Hindsight aliases `memory-retain`, `memory-reflect`, `memory-consolidate`
`hindsight`	Memory API (default host port 8888, `HINDSIGHT_PORT`)
`indexer`	Git, Obsidian, markdown trees, feed cache → Hindsight banks via `MEMORY_SINK=hindsight`

Health check after setup:

curl -fsS http://127.0.0.1:8888/health

Control plane URL is printed by the wizard (separate from the API port).

Secrets and config (never in the repo)

Path	Purpose
`~/.config/hindsight/hindsight.env`	Hindsight API settings
`~/.config/hindsight/litellm.yaml`	Optional full LiteLLM route override
`~/.config/hindsight/index.yml`	Index sources (index-sources.md)
`~/.config/hindsight/wizard.conf`	Active provider and wizard choices
`~/.hindsight/coding-agent.json`	Real file on disk — repo maps, opt-in paths, NotebookLM access

Shipped route templates live in config/litellm.lemond.yaml and config/litellm.ollama.yaml. Details: Model configuration.

Coding agents and MCP

Default (all providers): unified memory_* MCP plus in-repo session hooks:

npx personal-memory install-coding-agent --yes

Optional: also install the upstream Hindsight plugin hooks:

npx personal-memory install-coding-agent --yes --hindsight-plugin

MCP without session hooks:

npx personal-memory install-mcp --yes

Restart Cursor Agent after MCP changes.

Switch provider

npx personal-memory switch mem0 --yes
# return: npx personal-memory switch hindsight --yes

Updates compose profile, env, and MCP wiring. Does not migrate memories between backends.

NotebookLM (opt-in)

NotebookLM sync and MCP are provider-agnostic but were built for Hindsight-first workflows. See docs/notebooklm.md.