Skip to content

Local-first 🧠 memory for Hermes Agent that lives in your Obsidian vault and remembers project context. Nightly 😴 dream passes turn repeat corrections into confirmed preferences with measurable confidence. Adapters ship for Claude Code, Codex, and OpenClaw, with an MCP server for anything else.

TypeScript
1.6 KB
Raw Blame
# Open Second Brain β€” agent context for Aider

You have access to an Open Second Brain vault rooted at `{{VAULT}}`.
Open Second Brain keeps an observing-memory layer with your active
preferences, log, and recent retire events; reach it through the `o2b`
commands below, never by reading or editing its files.
`o2b store path` prints where that memory is stored.

You run as identity `@{{AGENT_NAME}}`.

## When to write to Brain

You cannot call Brain MCP tools from Aider β€” Aider has no MCP client at
the time this context was generated. Use the `o2b` CLI on PATH:

- `o2b brain feedback --topic <slug> --signal positive|negative \
   --principle "..." [--scope writing|coding|...]` when the user
  expresses a rule that should outlast the current task.
- `o2b brain apply-evidence --pref <pref-id> --artifact "[[path]]" \
   --result applied|violated|outdated` after producing a durable
  artifact that touches an active preference.
- `o2b brain note "..."` for durable narrative milestones that
  fit neither category (release shipped, PR merged, fact discovered).

## Read access

- `o2b brain query --preference <id>` β€” fetch one preference by id.
- `o2b brain query --topic <slug>` β€” fetch preferences by topic.
- `o2b brain digest` β€” read the latest activity digest.
- `o2b brain export --format llms-txt` - list every active
  preference.

## Sanity checks

- `o2b brain doctor` β€” runs schema + invariant lints over the brain memory.
- `o2b install --check --target aider` β€” verifies that this Aider
  install is recognised by Open Second Brain.

This file is regenerated on every `o2b install --target aider --apply`.
Do not hand-edit; changes will be overwritten.