AgentOps Center shows how to trace multi-agent systems with OpenTelemetry and SigNoz

·3 views

A multi-agent workflow can slow down or grow costly while every dashboard still looks healthy, which is the operational problem AgentOps Center targets. The project is a small GitHub repository built for a SigNoz hackathon, and the author reviewed its code and README without deploying it. It defines span attributes for agents, tools and model calls, reusing gen_ai.* keys where possible and adding its own keys for workflow grouping. Its setup exports traces and metrics over OTLP to a Collector that writes to ClickHouse for self-hosted SigNoz, and the author notes that only the exporter would need to change for another backend. The author flags unverified points, including evolving GenAI conventions, a custom cost attribute, a fixed 100% sampling rate and a MIT license claim with no LICENSE file found, and says the repository was not tested against other backends.

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)

Most Websites Fail AI Readiness Checks, Scan of 370,000 Domains Finds

A scan of more than 370,000 domains by the AI-Primed tool examined how well websites serve AI answer engines and autonomous agents. The analysis found that these agents, such as PerplexityBot and ClaudeBot, prioritize cost and speed over full browser rendering, so sites can fail in ways standard SEO audits miss. Common problems include content that only appears after client-side JavaScript runs, which fast HTTP fetches never execute. Security systems from providers like Cloudflare and AWS WAF also often block AI user-agents with 403 errors or CAPTCHA challenges. The authors recommend serving key content in server-rendered HTML, explicitly allowing designated AI bots, and adding an llms.txt file that summarizes a site for language models.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why Your Claude API Bill Is High: Nine Common Causes and Fixes

Claude API costs often rise because of hidden thinking tokens, which are billed as output even when not displayed, and can be reduced by adjusting the effort setting. Prompt caching may silently fail when the start of a prompt changes between requests, such as through timestamps or reordered keys, and short prompts fall below the minimum cacheable length. Long conversations are costly because the full history is resent on each turn, so caching earlier turns and keeping tool results brief can help. Some charges are avoidable, including fast mode, which doubles the price, and real-time requests that could run through the half-price Batch API. The article advises saving each response's usage block to see where spending occurs, and judging cost by finished tasks rather than by per-request price.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Fezola now connects mailboxes and accounts to Grok as a custom connector

Users can add Fezola to Grok by entering the server URL https://api.fezola.com/mcp at grok.com/connectors, with no header required. Grok then sends the user to Fezola's consent screen, where they choose which sources to share and authorize access. Once connected, Grok can use Fezola's tools to search and act on messages across linked mailboxes and accounts, with each result labeled by its source account. Fezola creates drafts rather than sending messages by default, and direct sending must be enabled per mailbox. The setup was tested on 15 November 2026 and relies on xAI's documentation and Fezola's code.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Google's Perspective API shuts down December 31, 2026, prompting migration

Google's Jigsaw is discontinuing its Perspective API, which already stopped accepting new users earlier this year. After December 31, 2026, the service will stop responding to all remaining users. Tools such as Coral's comment filter, the Discourse plugin, and Drupal modules rely on its endpoint and will be affected. The article, written by the team behind the alternative VigiMod API, explains how to redirect existing calls by changing the host and key while keeping the same request and response format.

0 comments Read more at DEV Community