langsmith-cli Update Lets Teams Archive AI Traces to S3, Cutting Long-Term Storage Costs

·4 views

Developers have merged a new archive workflow into langsmith-cli that allows teams to store LangSmith AI debugging traces in their own private Amazon S3 buckets instead of paying for LangSmith's extended-retention tier. LangSmith's extended retention costs 10 times more per trace than its base tier, with the price difference reaching tens of thousands of dollars monthly at high production volumes. The workflow runs two scheduled export passes — one at two days and a reconciliation pass at twelve days — to capture all traces before the base 14-day retention window closes, deduplicating records by run ID. Archived traces are stored as Parquet files and can be queried locally using DuckDB without requiring a LangSmith API key. The update also highlights a billing risk where LangSmith's online evaluators and automation rules can silently upgrade traces to extended retention by default, potentially inflating costs at scale.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

MCP C# SDK v2.0+ Drops Object Wrapper for Array Tool Outputs in 2026-07-28 Protocol

The MCP C# SDK changed how array and scalar tool outputs are structured when both client and server negotiate the 2026-07-28 protocol version. Under the older 2025-11-25 protocol, array results had to be wrapped in an object envelope with a 'result' key inside 'structuredContent', but the newer protocol allows the value itself to be returned directly. The stable SDK (v2.0.0 and later) handles backward compatibility automatically, emitting the legacy envelope for older clients without requiring a separate handler. Developers can verify the correct wire shape by running offline contract tests using two separate client-server sessions, one per protocol version, connected via in-process pipes. Relying solely on unit tests against C# return types is insufficient, as they cannot detect whether an unintended wrapper has been reintroduced at the transport boundary.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

JavaScript-Only Links Block AI Crawlers From Discovering Web Pages, 41-Day Test Finds

SEO engineer Vinicius Stanula ran a 41-day field experiment on a roughly 1,000-page test site to measure how different crawlers handle JavaScript-injected navigation links. Pages linked only through JavaScript were entirely missed by AI bots including GPTBot, ClaudeBot, and PerplexityBot, while Googlebot could follow such links but with declining coverage at greater depths. When all links were switched to plain HTML, GPTBot discovered 250 new pages within 48 hours and Bingbot found 89 more, confirming the earlier gap was caused by JavaScript rendering limitations. The findings highlight that content rendered for human visitors via client-side scripts can remain invisible to crawlers that do not execute JavaScript. Stanula shared the experiment on LinkedIn as a practical warning for organizations seeking consistent content discovery across both traditional search engines and AI-driven platforms.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Mind Elixir Introduces Plain Text Format for Writing Mind Maps Without a GUI

Mind Elixir, a mind mapping tool, has introduced a plain text format that lets users create mind maps using indented Markdown-style lists instead of a graphical interface. The format uses simple two-space indentation to define node hierarchy, making it accessible to anyone familiar with Markdown. Users can also assign IDs to nodes using a bracket syntax to create bidirectional or unidirectional links between non-adjacent nodes. Additional features include JSON-style inline styling for individual nodes and a summary node syntax that annotates groups of sibling nodes. The format is designed to work in any text editor and is compatible with AI-generated outlines or plain-text note libraries, unlike proprietary formats such as .xmind or .mm.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How GCC Compiles C Code for 32-bit RISC-V: A Deep Dive into RV32I Assembly

A technical exploration examines how GCC transforms C source code into RISC-V 32-bit (RV32I) assembly by disassembling compiled binaries and comparing them against the original C code. The study uses a standardized C test corpus covering global variables, arithmetic operations, function calls, and control flow structures to stress-test the toolchain. RISC-V RV32I provides 32 general-purpose registers and a minimalist base instruction set of just 40 fixed-width 32-bit instructions, making compiler output easier to analyze than on legacy 8-bit architectures. Unlike x86, RISC-V has no dedicated condition flag registers; instead, conditional branches directly compare two registers, reducing hidden pipeline dependencies. QEMU and GDB are used to simulate and step through execution, enabling precise observation of how the compiler handles register allocation, stack frames, and memory addressing.

0 comments Read more at DEV Community