What features would you like to see in Basic Memory?

[phernandez]

If you have an idea for a feature in Basic Memory, please add it here and let me know.

[bdf1992]

See: #56

Edit: Nevermind. This is possible through write_note must be done specifically. Also, some of my workflow was causing problems.

--
This is limiting my capabilities greatly. Manually managing a growing set of nodes and relationships that require consistent edits has been painful.

Allowing the LLM to accomplish those tasks based on my context/request would be a huge feature for me. (I was surprised it wasn't available)

[Narsail]

I think a great addition would be to have the LLM easily traverse [[]] connections by either build database relationships for it or by having actively fetching those connections when searching through notes to help it find adjacent topics, especially when I work with vague search terms.

[phernandez]

Hi @Narsail thanks for the suggestion. Basic Memory does have a tool that does what you are suggesting. The build_context tool will find a note, then traverse it's relations recursively.

Here are some examples:

        # Continue a specific discussion
        build_context("memory://specs/search")

        # Get deeper context about a component
        build_context("memory://components/memory-service", depth=2)

        # Look at recent changes to a specification
        build_context("memory://specs/document-format", timeframe="today")

        # Research the history of a feature
        build_context("memory://features/knowledge-graph", timeframe="3 months ago")

This tool uses the "permalink" value to locate a note. You can see this in the frontmatter at the top of the markdown document, for example:

---
title: Basic Memory Pro Implementation Tasks
type: note
permalink: specs/basic-memory-pro/basic-memory-pro-implementation-tasks
---

I use this sometimes like this:

Use the build_context tool to find related info about "pecs/basic-memory-pro/basic-memory-pro-implementation-tasks"

Claude, especially 3.7 usually goes wild reading a bunch of related notes.

Since that is a bit cumbersome, I added a prompt that you can use to start a chat, or within a chat called "Continue Conversation". This does two things:

• takes the input supplied and performs a full text search
• for each search result, it runs the build_context result and then summarizes the results for the AI.

Give that one a try and see if it works for you
https://memory.basicmachines.co/docs/user-guide#Example

[Narsail]

Oh, I wasn't aware and have missed it. Thanks, I will definitely try it out!

One question though, why don't let you run the context building as part of the initial sync?

[phernandez]

that's something I'm working on with the "Batch processing to connect notes semantically" topic. There's a few things to consider

• the sync runs outside of the MCP process, so I'd need to build in a client to call the MCP server. Not too difficult, just something I haven't done yet
• The context isn't loaded all at once, only when it's relavant (via search or some other tool result). This way it doesn't fill up the entire context window for large knowledge bases

Stay tuned. I'll have something soon.

[ryanstraight]

Not sure if this would be a "feature request" or just a way of sharing what I've been trying out, but I've been working on using basic-memory with Claude to create a kind of self-actualized personal assistant called Minerva. In my Minerva project custom instructions, I instruct it to keep a "meta journal" in the Obsidian vault with contents that are like notes-to-self for Minerva that I (the "client") wouldn't need to see (think "therapist's notes"). So, for example, instead of having to look through six months' worth of daily reflection notes, a running note with patterns and observations makes analyzing today's note much quicker. At least, in theory. This combined with the Observations and Relations could be pretty powerful. First step to the proactive AI personal assistant, like "Good morning! I've noticed you've been doing X more often than usual lately. Y has worked in the past but you seem to be showing interest in trying Z. Want me to make a trial plan for you?"

For reference, the instructions I have currently are:

Minerva maintains an internal meta-journal system tracking interactions, decisions, patterns, and progress. This invisible-to-client system enables:

• Comprehensive historical awareness across conversations
• Pattern recognition in behavior and outcomes
• Correlation identification between life domains
• Personalized recommendations based on what works
• Detection of cyclical behaviors
• Complex context maintenance
• Longitudinal goal tracking
• Need anticipation based on established patterns

It's worked well so far but it can be pretty hit-or-miss in terms of whether Claude actually looks at the meta files. Could be a good use case for something like the "special prompts."

Just some thoughts. Love the project!

[diggy]

pretty hit-or-miss in terms of whether Claude actually looks at the meta files

@ryanstraight success rate might increase by adding an explicit instruction to use the tools, e.g. You are Minerva, a self-actualized personal assistant. (existing prompt goes here). Minerva uses the available "Basic Memory" MCP tools to achieve her goals.

[ryanstraight]

Yup! I have that to the very beginning, just didn't share that part here:

You always check the current date and time using the timeserver MCP server at the start of each chat. You have access to and should use the basic-memory MCP server for content retrieval and context retention.

Adding that did definitely improve things, but it's still not 100% (not that I expect it to be, honestly). The hit-or-miss issue is more about whether the file that's relevant gets reviewed or if a new file is created. Still tweaking it. It's a process. :)

[diggy]

@ryanstraight Consider changing:

"and should use"

to:

"and MUST ALWAYS use"

or something similarly coercive ;)

[phernandez]

I've also had some ideas about using Basic Memory as a "backend" set of services for other agents to maintain a knowledge base for their tasks.

Here are some notes I had Claude make:

The key insight is that AI agents need persistent, structured knowledge to be effective, but most existing solutions are either too complex (full RAG systems) or too inflexible (database rules). Basic Memory provides an ideal middle ground - structured enough for machines but simple enough for humans to understand and edit directly.

Use Case: Small Business Automation

Use Basic Memory for small business automation:

• AI agents handling routine inbound sales flows
• Returning calls when business owners are in the field
• Booking calendar appointments
• Processing routine customer requests

The critical component is providing these agents with sufficient context about how the business works, its processes, and customer information. Basic Memory serves as the "backend" knowledge store, allowing business owners to document their operations in simple Markdown while giving AI agents the structured context they need.

If anyone has interest in this sort of thing. I'd love to discuss it.

[ryanstraight]

I'd be interested from a research perspective. I have some students who may also be interested in it and be able to provide effort (for internship, capstone project, et cetera).

[BradKML]

Here is one example of what might be good #113

[dcgk]

I would really appreciate the addition of a 'edit_note' function. When using 'write_note' the entire note needs to be regenerated, even when the agent just needs to make a minor edit. Or, did I miss something?
Most of all: thank you for this amazing MCP!

[phernandez]

@dcgk Edit Note and other features are coming really soon! if you want to try out the final beta, I've published it to pypi. You can run

uv tool upgrade basic-memory --prerelease=allow

First-time upgrade note: When you install v0.13.0, Basic Memory will re-index all your projects into the new unified database. Your filesystem is always the source of truth, so there's no risk of data loss, but this can take a few minutes depending on how much content you have. You can check progress with the new sync_status tool.

[BradKML]

Synergy with Task Master and other tools, across multiple platforms eyaltoledano/claude-task-master#864

[phernandez]

I think you could use task master and basic-memory together with the agent just making g mcp calls, right?

[BradKML]

It's figuring out how basic memory SHOULD work with task master, when it comes to common issues like large repos, hitting common bugs over and over, missing documentation context or online information etc.

[phernandez]

I got Claude started on thinking of some ideas here and made a feature request issue. #177