memory-notes

Memory Notes

Write well-structured notes that Basic Memory can parse into a searchable knowledge graph. Every note is a markdown file with three key sections: frontmatter, observations, and relations.

Note Anatomy

---
title: API Design Decisions
tags: [api, architecture, decisions]
---

# API Design Decisions

The API team evaluated multiple approaches for the public API during Q1. After
prototyping both REST and GraphQL, the team chose REST due to broader ecosystem
support and simpler caching semantics. This note captures the key decisions and
their rationale, along with open questions still to resolve.

## Observations
- [decision] Use REST over GraphQL for simplicity #api
- [requirement] Must support versioning from day one
- [risk] Rate limiting needed for public endpoints

## Relations
- implements [[API Specification]]
- depends_on [[Authentication System]]
- relates_to [[Performance Requirements]]

Frontmatter

Every note starts with YAML frontmatter:

---
title: Note Title          # required — becomes the entity name in the knowledge graph
tags: [tag1, tag2]         # optional — for organization and filtering
type: note                 # optional — defaults to "note", use custom types with schemas
permalink: custom-path     # optional — auto-generated from title if omitted
---

• The title must match the # Heading in the body
• Tags are searchable and help with discovery
• Custom type values (Task, Meeting, Person, etc.) work with the schema system. See the memory-schema skill for defining schemas, validating notes against them, and detecting drift.
• The permalink is auto-generated from the title and directory. For example, title "API Design Decisions" in directory "specs" produces permalink specs/api-design-decisions and memory URL memory://specs/api-design-decisions. If no directory is specified, the permalink is just the kebab-cased title. By default permalinks stay stable across file moves (unless the project enables update_permalinks_on_move). You rarely need to set one manually.

Note: bm_write accepts an exact file path as identifier, or title + directory. content may include YAML frontmatter; alternatively, supply title, tags, note_type, and metadata separately and let Basic Memory generate it. For an overwrite, use the file_path returned by bm_read as identifier and pass its checksum as expected_checksum.

Body / Context

Free-form markdown between the heading and the Observations section. This is the heart of the note — write generously here:

• Background, motivation, and history
• Detailed explanation of what happened and why it matters
• Analysis, reasoning, and trade-offs considered
• Context that someone (or an AI) needs to understand this note later

Write complete, substantive prose. Basic Memory's search retrieves relevant chunks from note bodies, so longer, richer context makes notes more discoverable and more useful when found. Don't reduce everything to bullet points — tell the story.

Observations

Observations are categorized facts — the atomic units of knowledge. Each one becomes a searchable entity in the knowledge graph.

Syntax

- [category] Content of the observation #optional-tag

• Square brackets define the semantic category
• Content is the fact, decision, insight, or note
• Hash tags (optional) add extra metadata for filtering

Categories Are Arbitrary

The category in brackets is free-form — use whatever label makes sense for the observation. There is no fixed list. The only rule is the [category] content syntax. Consistency within a project helps searchability, but invent categories freely.

A few examples to illustrate the range:

- [decision] Use PostgreSQL for primary data store
- [risk] Third-party API has no SLA guarantee
- [technique] Exponential backoff for retry logic #resilience
- [question] Should we support multi-tenancy at the DB level?
- [preference] Use Bun over Node for new projects
- [lesson] Always validate webhook signatures server-side
- [status] active
- [flavor] Ethiopian beans work best with lighter roasts

Observation Tips

• One fact per observation. Don't pack multiple ideas into one line.
• Be specific. [decision] Use JWT is less useful than [decision] Use JWT with 15-minute expiry for API auth.
• Use tags for cross-cutting concerns. [risk] Rate limiting needed #api #security makes this findable under both topics.
• Categories are queryable. bm_search(items=["observation"], categories=["decision"]) returns every decision observation across your knowledge base.

When a Claim Holds

An observation can say when it is true. Put a qualifier after the category:

- [decision] @effective:2026-06-10 The cache layer uses Redis
- [event] @occurred:2026-10-14 Launch sync moved the date to Thursday
- [deadline] @due:2026-10-20 Draft ready for review
- [stage] @valid:2026-10 Series B, about 200 employees
- [decision] @effective[2026-06-10,2026-10-14) Launch on Friday

• Kinds: effective (a decision or rule takes effect), occurred (an event happened), due (a commitment), mentioned (when it was said), and valid, the default (@2026-06-10 means @valid:2026-06-10).
• Points cover the span their precision names: @2026 is that year, @2026-06 that month, and @2026-06-10 from that date onward.
• Ranges use [start,end), and either end can be open: @effective[2026-06-10,).
• Dates are written out. An unquoted date is one token (@occurred:2026-06-10). Quote anything longer: @occurred:"June 10, 2026". A relative date like yesterday is never read as a qualifier; it stays in the text.
• When a fact changes, close the old one's range and start the new one, instead of deleting the old one. Both stay searchable by date.

Search by what notes say was true, rather than by when they were edited:

bm_search(valid_at="2026-07-01")                              # true on that date
bm_search(time_kind="due", valid_overlaps="[2026-10-14,2026-10-21)")  # due this week

A valid-time filter returns only dated observations; drop it to search everything. Day points run "from that date onward", so a due-date window also returns items that are already overdue.

Relations

Relations create edges in the knowledge graph, linking notes to each other. They're how you build structure beyond individual notes.

Syntax

- relation_type [[Target Note Title]]

• relation_type is a descriptive verb or phrase (snake_case by convention)
• Double brackets [[...]] identify the target note by title or permalink
• Relations are directional: this note → target note

Relation Types

Type	Purpose	Example
implements	One thing implements another	- implements [[Auth Spec]]
requires	Dependencies	- requires [[Database Setup]]
relates_to	General connection	- relates_to [[Performance Notes]]
part_of	Hierarchy/composition	- part_of [[Backend Architecture]]
extends	Enhancement or elaboration	- extends [[Base Config]]
pairs_with	Things that work together	- pairs_with [[Frontend Client]]
inspired_by	Source material	- inspired_by [[CRDT Research Paper]]
replaces	Supersedes another note	- replaces [[Old Auth Design]]
depends_on	Runtime/build dependency	- depends_on [[MCP SDK]]
contrasts_with	Alternative approaches	- contrasts_with [[GraphQL Approach]]

Inline Relations

Wiki-links anywhere in the note body — not just the Relations section — also create graph edges:

We evaluated [[GraphQL Approach]] but decided against it because
the team has more experience with REST. See [[API Specification]]
for the full contract.

These create references relations automatically. Use the Relations section for explicit, typed relationships; use inline links for natural prose references.

Relation Tips

• Link liberally. Relations are what turn isolated notes into a knowledge graph. When in doubt, add the link.
• Create target notes if they don't exist yet. [[Future Topic]] is valid — BM will resolve it when that note is created.
• Use bm_context to traverse. bm_context(identifier="memory://note-title") follows relations to gather connected knowledge.
• Custom relation types are fine. taught_by, blocks, tested_in — use whatever is descriptive.

Memory URLs

Every note is addressable via a memory:// URL, built from its permalink. These URLs are how you navigate the knowledge graph programmatically.

URL Patterns

memory://api-design-decisions          # by permalink (title → kebab-case)
memory://docs/authentication           # by file path
memory://docs/authentication.md        # with extension (also works)
memory://auth*                         # wildcard prefix
memory://docs/*                        # wildcard suffix
memory://project/*/requirements        # path wildcards

Project-Scoped URLs

In multi-project setups, prefix with the project name:

memory://main/specs/api-design         # "main" project, "specs/api-design" path
memory://research/papers/crdt          # "research" project

The first path segment is matched against known project names. If it matches, it's used as the project scope. Otherwise pass project explicitly when several projects are addressable; unqualified calls are refused.

Using Memory URLs

Memory URLs work with bm_context to assemble related knowledge by traversing relations:

# Get a note and its connected context
bm_context(identifier="memory://api-design-decisions")

# Wildcard — gather all docs
bm_context(identifier="memory://docs/*")

# Direct read by permalink
bm_read(identifier="memory://api-design-decisions")

Before Creating a Note

Always search Basic Memory before creating a new note. Duplicates fragment your knowledge graph — updating an existing note is almost always better than creating a second one.

Search with Multiple Variations

A single search often misses. Try the full name, abbreviations, acronyms, and keywords:

# Searching for an entity that might already exist
bm_search(query="Kubernetes Migration")
bm_search(query="k8s migration")
bm_search(query="container migration")

For people, try full name and last name. For organizations, try the full name and common abbreviations.

Decision Tree

• Entity exists → Update it with bm_edit (append observations, add relations, find-and-replace outdated info)
• Entity doesn't exist → Create it with bm_write
• Unsure if it's the same entity → Read the existing note first, then decide

Granular Updates with bm_edit

When a note already exists, make targeted edits instead of rewriting the whole file:

# Add a new observation under the existing Observations heading
bm_edit(
  identifier="API Design Decisions",
  command="insert_after_section",
  section="Observations",
  content="- [decision] Switched to OpenAPI 3.1 for spec generation #api"
)

# Fix outdated information
bm_edit(
  identifier="API Design Decisions",
  command="find_replace",
  find_text="- [status] draft",
  content="- [status] approved"
)

# Add a new relation
bm_edit(
  identifier="API Design Decisions",
  command="insert_after_section",
  section="Relations",
  content="- depends_on [[Rate Limiter]]"
)

This preserves existing content and keeps the edit history clean.

Writing Notes with Tools

Creating a Note

bm_write(
  title="API Design Decisions",
  directory="architecture",
  tags=["api", "architecture"],
  content="""# API Design Decisions

The API team evaluated REST and GraphQL during Q1 planning. After prototyping
both approaches, we chose REST for the public API — broader ecosystem support,
simpler caching with HTTP semantics, and a lower learning curve for external
consumers. GraphQL remains an option for internal services where query
flexibility matters more.

## Observations
- [decision] Use REST for public API #api
- [requirement] Support API versioning from v1

## Relations
- implements [[API Specification]]
- relates_to [[Backend Architecture]]"""
)

Basic Memory auto-generates frontmatter (including the permalink and memory URL) from the parameters. This note would get permalink architecture/api-design-decisions and be addressable at memory://architecture/api-design-decisions.

Editing an Existing Note

Use bm_edit to update a note in place — seven commands: append, prepend, find_replace, replace_section, insert_before_section, insert_after_section, and set_metadata (merge frontmatter without changing the body).

# insert_after_section / insert_before_section — add lines at an existing heading
bm_edit(
  identifier="API Design Decisions",
  command="insert_after_section",
  section="Observations",
  content="- [decision] Use OpenAPI 3.1 for spec generation #api"
)

# append / prepend — add to the end of the note or the start of the body
# (use for time-ordered logs; neither one targets a section)
bm_edit(
  identifier="API Design Decisions",
  command="prepend",
  content="> Updated 2026-05-28: auth approach finalized.\n"
)

# replace_section — rewrite a named section (use for living content that stays current)
bm_edit(
  identifier="API Design Decisions",
  command="replace_section",
  section="Summary",
  content="Concise, current summary of the decision and its rationale."
)

# find_replace — swap specific text
bm_edit(
  identifier="API Design Decisions",
  command="find_replace",
  find_text="OpenAPI 3.0",
  content="OpenAPI 3.1"
)

When an edit is destructive (replace_section, find_replace), read the note first and pass its checksum, so the edit applies only to the revision you read:

note = bm_read(identifier="API Design Decisions", output_format="json")
bm_edit(
  identifier="API Design Decisions",
  command="replace_section",
  section="Summary",
  content="Concise, current summary of the decision and its rationale.",
  expected_checksum=note["checksum"]
)

A checksum conflict means someone changed the note after you read it. Read it again and redo the edit against the new text. bm_write(overwrite=True) takes expected_checksum the same way.

Moving a Note

Use bm_move to reorganize notes into different directories:

bm_move(
  identifier="API Design Decisions",
  destination="archive/api-design-decisions.md"
)

By default the permalink stays the same after a move, so links keep resolving. Projects with update_permalinks_on_move enabled rewrite it from the new path.

Best Practices

1. Start with context. Before listing observations, explain why this note exists. Future-you (or your AI collaborator) will thank you.

2. Favor completeness. Write rich, substantive notes. Basic Memory's search pulls relevant chunks from note bodies, so longer notes with more context are more discoverable, not less. Use prose in the body to tell the full story — the background, the reasoning, the nuance. Then distill key facts into [category] content observations for structured queries. Both matter: prose gives meaning, observations give precision.

3. Build incrementally. Add to existing notes rather than creating duplicates. Use bm_edit to append new observations or relations as you learn more.

4. Use consistent titles. Note titles are identifiers in the knowledge graph. API Design Decisions and Api Design decisions are different entities. Pick a convention and stick with it.

5. Link related concepts. The value of a knowledge graph compounds with connections. A note with zero relations is an island — useful, but not as powerful as a connected one.

6. Let the graph grow naturally. Don't try to design a perfect taxonomy upfront. Write notes as you work, add relations as connections emerge, and periodically use the memory-reflect or memory-defrag skills to consolidate.