Save your AI conversations to a notes base

Every Base76 account starts with a base called notes. There’s no schema to design and nothing to name. Tell your agent to save something and it calls save_note. Next month, in a different chat or a different agent, ask what you decided and it searches.

The prompt

Save the decisions from this conversation to my notes, tagged decisions.

That’s the whole setup. The agent writes a markdown note with a title, the body, your tags, and a source line saying where it came from. If it skips the title, Base76 takes the first heading or line of the body.

What’s in the base

Field	Type	Notes
title	text	Display field
body	markdown	Up to 200,000 characters per note
tags	multi_select	New tags are added as you use them
source	text	The conversation topic, a URL, or a script name

Find it later

Ask “what did we pick for the kitchen?” and the agent runs search. Search mixes keyword matching with semantic matching, so it finds a note even when your question uses different words than the note does. Keyword results include a new note right away. Semantic matches follow about two minutes after a write.

curl -s https://api.base76.com/v1/bases/tim/notes/search \
  -H "Authorization: Bearer $B76_KEY" \
  -H 'content-type: application/json' \
  -d '{"query": "what did we pick for the kitchen", "limit": 3}'

Each hit has the collection, the record id, a title, a score, and a snippet with the matching words marked. Pass "include_records": true to get the full notes back in the same call.

Save from a script

save_note is also a plain REST endpoint, so a deploy script or a cron job can keep a log without any setup:

curl -s https://api.base76.com/v1/notes \
  -H "Authorization: Bearer $B76_KEY" \
  -H 'content-type: application/json' \
  -d '{"body": "# Deployed v2.3\n\nMigrations ran clean.",
       "tags": ["deploys"], "source": "deploy.sh"}'

The response includes the saved record, the base it went to, and its audience: "audience": "private".

Hand a note to another agent

Share a note and you get a read-only link. Add .md to the end and the link returns the note as markdown with frontmatter: the title, tags, source, and dates, followed by the body. Paste that URL into any agent that can fetch a page and it reads the note directly.

Let it grow

Notes often turn into something with structure, like a reading list or a set of meeting decisions with owners. Ask the agent to add fields or a new collection. Adding is immediate. Removing a field or changing its type comes back as confirm_required with the number of records affected, and the agent has to show you that before it confirms.

Limits

• save_note always writes to your own notes base. To save into a shared base, the agent uses create on that base instead.
• The notes base can’t be deleted.
• Semantic search trails writes by about two minutes. Keyword search is current.