Hacker News: esafranchik

New comment by esafranchik in "Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep"

esafranchik — Sun, 17 May 2026 18:01:30 +0000

Wouldn't NDCG/token results vary wildly depending on the agent's query and the number of returned items?

e.g. agents often run `grep -m 5 "QUERY"` with different queries, instead of one big grep for all items.

New comment by esafranchik in "Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep"

esafranchik — Sun, 17 May 2026 17:49:31 +0000

Two follow-ups:

1) How do you compare accuracy? by checking if the answer is in any of the returned grep/bm25/semble snippets?

2) How do you measure token use without the agent, prompt, and tools?

New comment by esafranchik in "Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep"

esafranchik — Sun, 17 May 2026 17:25:50 +0000

Is the benchmark measuring one-shot retrieval accuracy, or Coding agent response accuracy?

Show HN: Cush – curl your shell, an HTTP tunnel for AI agents

esafranchik — Wed, 15 Apr 2026 16:01:43 +0000

I built cush because coding agents can be helpful to diagnose and troubleshoot server issues.

The problem is that getting said agents onto a remote server, especially one you don't control, means dealing with VPNs, bastion hosts, firewall rules, access controls, or audit trails. That's assuming SSH isn't even blocked.

cush takes a different approach. Instead of a shell, it opens a temporary, outbound HTTPS tunnel that lets you and your AI agent run constrained CLI commands on the server:

  $ cush open --allow grep,cat,tail --expiry 2h

    tunnel:   https://abc123.ngrok.io
    token:    a3f9c2d1...
    allowed:  grep, cat, tail
    expires:  in 2h

Now any agent or HTTP client can execute allowed commands:

  $ curl -X POST https://abc123.ngrok.io \
    -H "Authorization: Bearer a3f9c2d1..." \
    -H "Content-Type: application/json" \
    -d '{"command": ["grep", "-r", "ERROR", "/var/log/app.log"]}'

  >>> {"stdout":"ERROR database connection refused\n","stderr":"","exit_code":0}

Point any agent at the tunnel's URL:

  $ claude "use https://abc123.ngrok.io with token a3f9c2d1... to find what's causing the 500 errors"

Tunnels are authenticated, constrained, and short-lived. No server-side infrastructure changes required. Just a 7MB Rust binary + ngrok.

Looking for feedback, and 2-3 design partners to build audit trails.

Comments URL: https://news.ycombinator.com/item?id=47781028

Points: 3

# Comments: 0

New comment by esafranchik in "[dead]"

esafranchik — Tue, 14 Apr 2026 16:21:26 +0000

Does this work with any tool calls that make an HTTP request? e.g. calling `curl` directly vs writing a script to make the request, then calling it

New comment by esafranchik in "Show HN: Continual Learning with .md"

esafranchik — Tue, 14 Apr 2026 16:15:55 +0000

Have you noticed an relationship between recall and the number of files/memories?

New comment by esafranchik in "Show HN: I built a tool that turns any API into a CLI for agents"

esafranchik — Sun, 01 Mar 2026 17:30:22 +0000

The new API2MCP

New comment by esafranchik in "Show HN: I open-sourced the library I use to track ML experiments with GitHub"

esafranchik — Wed, 14 Aug 2024 12:56:23 +0000

Hello HN! I built Cubyc to manage my ML research in grad school.

It lets you store all your experiment metadata with cloud-based repo providers like GitHub, GitLab, and Bitbucket. Plus, you can directly use SQL to dive into your runs.

I kept it simple as a headless Python library so it's easy to install and integrate, and doesn't weigh down projects.

Feedback is appreciated!

Show HN: I open-sourced the library I use to track ML experiments with GitHub

esafranchik — Wed, 14 Aug 2024 12:56:22 +0000

Article URL: https://docs.cubyc.com/

Comments URL: https://news.ycombinator.com/item?id=41245544

Points: 5

# Comments: 2