This is a proposal, not a bug report. Happy to discuss and contribute PRs
for any of the points below if the direction looks good to maintainers.
Context
I went through all the docs pages (getting-started, architecture, python-api,
cli, faq, claude-plugin) and collected a list of concrete improvements that
would make memsearch easier to adopt for new users and easier to maintain
going forward.
Proposed improvements
1. Fix broken Integrations page (404)
mkdocs.yml has Integrations: integrations.md in the nav, but the file
does not exist. Anyone clicking that link on the docs site gets a 404.
Suggestion: either create a minimal integrations.md with LangChain /
LangGraph / LlamaIndex / CrewAI examples (the README already has stubs for
these), or remove the nav entry until the page is ready.
2. Add a zero-config quick-start
The most common entry point for new users — no API key, local embeddings,
local Milvus Lite — requires reading across several pages to get working.
A single self-contained section ("Up and running in 2 minutes, no API key
needed") would lower the barrier significantly:
pip install "memsearch[local]"
mem = MemSearch(paths=["./notes"], embedding_provider="local")
await mem.index()
results = await mem.search("your query")
3. Add platform support / requirements section
After issues #89 and #90 it is clear that Windows support is a frequent
question. Currently there is no single place in the docs that states which
platforms are supported. Suggested addition to the Installation section:
Platform support
Milvus Lite (the default backend) runs on Linux and macOS only.
On Windows, use Milvus Server via Docker or Zilliz Cloud — see
Milvus Backends.
4. Expand the FAQ
Currently only 3 questions. Suggested additions based on common issues:
| Question |
Answer summary |
| Does memsearch work on Windows? |
Not with Milvus Lite — use Docker or WSL2 |
| How do I wipe and rebuild the index? |
memsearch reset --yes && memsearch index |
| How do I see what is indexed? |
memsearch stats |
| Why is search returning irrelevant results? |
Check embedding provider / try re-indexing |
| What is "dimension mismatch" and how do I fix it? |
memsearch reset --yes when switching providers |
5. Reduce duplication between docs/claude-plugin.md and ccplugin/README.md
Both files cover the same topics (architecture, quick start, troubleshooting).
When one is updated, the other goes stale. Possible approaches:
- Make
ccplugin/README.md the canonical source for plugin users (people who
clone the repo), and have docs/claude-plugin.md focus on concepts and
link to it
- Or consolidate everything into
docs/claude-plugin.md and make
ccplugin/README.md a short pointer
6. Split docs/claude-plugin.md into sub-pages
At 930+ lines it covers too much in one place. Suggested split:
claude-plugin/overview.md — what it is, comparison table, quick start
claude-plugin/how-it-works.md — hooks lifecycle, progressive disclosure
claude-plugin/configuration.md — config options, collection naming
claude-plugin/troubleshooting.md — known issues and fixes
7. Add cross-links between pages
A few obvious missing links:
python-api.md → architecture.md when mentioning chunk_hash and
composite IDs
cli.md → getting-started.md for first-time usage context
getting-started.md → faq.md at the bottom ("Common questions →")
8. Add a troubleshooting section outside the plugin docs
Common errors (dimension mismatch, connection refused, API key missing) are
currently buried inside the plugin troubleshooting section. A top-level
Troubleshooting page (or section in FAQ) would help users of the Python
library and CLI who never use the plugin.
Summary
| # |
Change |
Effort |
Impact |
| 1 |
Fix Integrations 404 |
Low |
Medium |
| 2 |
Zero-config quick-start |
Low |
High |
| 3 |
Platform support section |
Low |
High |
| 4 |
Expand FAQ |
Low |
High |
| 5 |
Deduplicate plugin docs |
Medium |
Medium |
| 6 |
Split claude-plugin.md |
Medium |
Medium |
| 7 |
Add cross-links |
Low |
Medium |
| 8 |
Troubleshooting page |
Medium |
Medium |
Context
I went through all the docs pages (getting-started, architecture, python-api,
cli, faq, claude-plugin) and collected a list of concrete improvements that
would make memsearch easier to adopt for new users and easier to maintain
going forward.
Proposed improvements
1. Fix broken Integrations page (404)
mkdocs.ymlhasIntegrations: integrations.mdin the nav, but the filedoes not exist. Anyone clicking that link on the docs site gets a 404.
Suggestion: either create a minimal
integrations.mdwith LangChain /LangGraph / LlamaIndex / CrewAI examples (the README already has stubs for
these), or remove the nav entry until the page is ready.
2. Add a zero-config quick-start
The most common entry point for new users — no API key, local embeddings,
local Milvus Lite — requires reading across several pages to get working.
A single self-contained section ("Up and running in 2 minutes, no API key
needed") would lower the barrier significantly:
pip install "memsearch[local]"3. Add platform support / requirements section
After issues #89 and #90 it is clear that Windows support is a frequent
question. Currently there is no single place in the docs that states which
platforms are supported. Suggested addition to the Installation section:
4. Expand the FAQ
Currently only 3 questions. Suggested additions based on common issues:
memsearch reset --yes && memsearch indexmemsearch statsmemsearch reset --yeswhen switching providers5. Reduce duplication between
docs/claude-plugin.mdandccplugin/README.mdBoth files cover the same topics (architecture, quick start, troubleshooting).
When one is updated, the other goes stale. Possible approaches:
ccplugin/README.mdthe canonical source for plugin users (people whoclone the repo), and have
docs/claude-plugin.mdfocus on concepts andlink to it
docs/claude-plugin.mdand makeccplugin/README.mda short pointer6. Split
docs/claude-plugin.mdinto sub-pagesAt 930+ lines it covers too much in one place. Suggested split:
claude-plugin/overview.md— what it is, comparison table, quick startclaude-plugin/how-it-works.md— hooks lifecycle, progressive disclosureclaude-plugin/configuration.md— config options, collection namingclaude-plugin/troubleshooting.md— known issues and fixes7. Add cross-links between pages
A few obvious missing links:
python-api.md→architecture.mdwhen mentioningchunk_hashandcomposite IDs
cli.md→getting-started.mdfor first-time usage contextgetting-started.md→faq.mdat the bottom ("Common questions →")8. Add a troubleshooting section outside the plugin docs
Common errors (dimension mismatch, connection refused, API key missing) are
currently buried inside the plugin troubleshooting section. A top-level
Troubleshootingpage (or section in FAQ) would help users of the Pythonlibrary and CLI who never use the plugin.
Summary