RepoJournal

Developer journal

Anurag Verma

@anurag629 since Jan 2022

144 stories across 32 days in 2021

Mon Wed Fri
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
an entry to read activity only Less More before entries began Recent weeks →

Entries start Jan 13, 2022 — we reconstructed the previous month from @anurag629's public GitHub activity on the day they joined. Squares before that show contribution counts only.

Add this graph to your GitHub README

Paste this Markdown into your README — the badge stays current automatically and links back here.

[![anurag629's work on RepoJournal](https://repojournal.com/journals/anurag629/heat.svg)](https://repojournal.com/journals/anurag629)

Friday, Jul 24, 2026

TUI slash commands and AWS Bedrock provider for luca-py

Anurag Verma shipped interactive command pickers to the terminal UI and added AWS Bedrock as the third provider, testing the transport abstraction against a fundamentally different API shape.

The main work landed in quantized-ai/luca-py across three PRs. First, Verma added TUI slash commands [1] with model and reasoning pickers, all contained in contrib/tui without touching core or client code. The /help command lists every registered command rendered from the registry so it never drifts [2], and /model opens a hierarchical picker to select a provider first, then one of its models, applied together so a provider is never left with a mismatched model.

Second, Verma implemented AWS Bedrock as the third real provider via the Converse API [3]. Unlike the OpenAI and Anthropic providers already in the codebase, Bedrock puts the model id in the URL path rather than the request body, uses binary framing for streaming instead of server-sent events, and fronts multiple model families through a single endpoint. Verma translated the internal block types to the Converse payload format and hoisted the system prompt to a top-level array.

Third, he fixed two display issues in the TUI [4]. Assistant and thinking cells now render markdown via rich.markdown.Markdown instead of showing literal asterisks and backticks, with a _renderable() hook on TranscriptCell handling the transformation while streaming deltas stay plain. He also clipped long tool-call arguments so they do not break the display layout.

Sources

  1. Add TUI slash commands with model/reasoning pickers (#6) · quantized-ai/luca-py
  2. Add TUI slash commands with model/reasoning pickers · quantized-ai/luca-py
  3. Add AWS Bedrock provider via the Converse API (#5) · quantized-ai/luca-py
  4. Render markdown in cells and clip tool-call args in the TUI (#7) · quantized-ai/luca-py

On this day

← All entries

Follow along

Follow anurag629

A short briefing every day anurag629 ships something — in about 3 minutes.

or

One email a day. Unsubscribe in one click. Read a past issue →

RSS Read other developers