> Docker `:latest` tag guarantees you always get current docs without manual updates.
The docs should probably be pinned to the version of the tool you have installed. Aside from that, pinning to a specific container hash (not tag) allows you to audit it and trust that malicious instructions haven’t been introduced into the docs.
Thank you so much for pointing that out, I just updated the docker instructions in the README and on DockerHub:
```shell
# Pin to commit hash for production security
# Get current hash from: https://hub.docker.com/r/keminghe/py-dep-man-companion/tags
docker pull keminghe/py-dep-man-companion@sha256:2c896dc617e8cd3b1a1956580322b0f0c80d5b6dfd09743d90859d2ef2b71ec6 # 2025-07-22 release example
# Or use latest for development
docker pull keminghe/py-dep-man-companion:latest
```
I currently have poor internet connection and I am not able to view the demo, but having read the description of the project I failed to understand what it is/does and which problems/tasks it solves.
This is great feedback. I just improved the README and DockerHub overview to be more clear:
"Stop getting out-of-date Python package manager commands from your AI. Cross-reference latest official pip, poetry, uv, and conda docs with auto-updates."
The demo is not convincing, I rarely find myself migrating between package managers and if I do I would expect claude code to ace this task without mcp help.
Appreciate the feedback. I will make it my todo to try out your suggestion of comparing Claude Code with and without MCP to measure the quantitative difference.
Even though your demo is not that helpful, I appreciate you sharing this. I think it opens my eyes for another ideas of providing better documentation for coding agent. I believe the current RAG-based approach for coding agent is not the most optimal solution.
Thank you. Yes, 50% of the core value prop is the self-updating automation and the traditional fuzzy + full text search capabilities that disrupt the embedding-centric RAG paradigm. Plus, Tantivy (Rust-based) is fast.
I think in terms of providing the widest coverage, I think Dash[0] which has been in the offline documentation space for a long time should have everyone beat.
Appreciate you opening my eyes to this. Dash is indeed comprehensive, and a much bigger initiative. I wonder how it's handling the documentation staleness issue? New docs are published every minute.
For the package manager ecosystem it supports, it relies on the projects auto-generated docs, and the builds the docsets from those. I guess it does that step in a cached on-demand way. That way, it can provide docs to all the packages and package versions.
E.g. for Rust: Crate is published crates.io -> triggers automatic docs build on docs.rs -> Dash clients can now pull docsets through a proxy that builds the docsets built on the static HTML bundles built for docs.rs.
I see, yes, with this implementation, the networking component becomes necessary. I made different design decisions for `py-dep-man-companion` such that the tool is fully offline and can be used with local llms.
The docs should probably be pinned to the version of the tool you have installed. Aside from that, pinning to a specific container hash (not tag) allows you to audit it and trust that malicious instructions haven’t been introduced into the docs.
```shell # Pin to commit hash for production security # Get current hash from: https://hub.docker.com/r/keminghe/py-dep-man-companion/tags docker pull keminghe/py-dep-man-companion@sha256:2c896dc617e8cd3b1a1956580322b0f0c80d5b6dfd09743d90859d2ef2b71ec6 # 2025-07-22 release example
# Or use latest for development docker pull keminghe/py-dep-man-companion:latest ```
"Stop getting out-of-date Python package manager commands from your AI. Cross-reference latest official pip, poetry, uv, and conda docs with auto-updates."
[0]: https://blog.kapeli.com/dash-8
E.g. for Rust: Crate is published crates.io -> triggers automatic docs build on docs.rs -> Dash clients can now pull docsets through a proxy that builds the docsets built on the static HTML bundles built for docs.rs.