How to Build an AI Support Copilot That Separates Runbooks from Tool Access

·1 views

A tutorial published on DEV Community outlines how to build a structured AI support copilot that separates investigative instructions from live system access. The design uses two distinct layers: runbooks that guide problem investigation and narrowly scoped tools that interact with live systems via a common protocol. Rather than loading all procedures into every conversation, the workflow routes each support request to a relevant runbook only when needed, reducing context bloat and preventing irrelevant instructions from interfering. Authorization decisions are handled in code, not delegated to the AI model, ensuring the model can propose tool calls but never independently approve them. The goal is faster human-assisted investigation, not autonomous resolution, keeping operational authority firmly with human agents.

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)

Data Privacy Advocate Joseph Sides Joins DEV to Demystify Online Tracking

Joseph Sides, a Florida-based data privacy advocate and consultant, has introduced himself to the DEV Community as a new contributor. With a background in criminology, business, and data analytics, Sides developed his focus on privacy after growing concerned about the gap between what users see online and how their data is actually collected and shared. He argues that tracking technologies such as advertising pixels, cookies, and analytics tools often operate without meaningful consumer awareness or understanding. On DEV, he plans to publish accessible content covering how website tracking works, why consent and transparency matter, and how individuals can better protect their personal information. Sides believes strong privacy practices and business innovation are not mutually exclusive, and that consumer trust depends on companies being genuinely transparent about their data handling.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Istio Moves Networking Code Out of Microservices and Into Infrastructure

Splitting a monolith into many microservices forces each service to carry repetitive networking code — such as TLS, retries, timeouts, and tracing — that is unrelated to core business logic. Maintaining this through shared SDKs becomes unsustainable when multiple programming languages are involved, as each language requires its own SDK that can drift in behavior over time. A service mesh like Istio addresses this by shifting all such concerns into the infrastructure layer, automating certificate management, enforcing retry policies via YAML, and providing built-in observability through tools like Prometheus and Jaeger. However, the approach carries real trade-offs, including approximately 60MB and 0.2 vCPU overhead per sidecar and measurably higher latency compared to ambient mode. The article argues that a service mesh is best suited for multi-language environments with mature operational teams, and provides a decision framework for cases where it may not be appropriate.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Claude on Mobile: Anthropic's Channels vs Open-Source Telechat Compared

Users wanting Claude AI on their phones now have two distinct options: Claude Code Channels, Anthropic's official cloud-based integration, and Telechat, an open-source self-hosted alternative built by an independent developer. Channels connects via Telegram or Discord and routes messages through Anthropic's servers, requiring a Claude Pro subscription starting at $20 per month with zero infrastructure setup. Telechat runs locally on a user's own machine, supports more platforms including WhatsApp and Slack, and charges per API token rather than a flat subscription fee. A smart model-routing feature in Telechat automatically selects the most cost-effective Claude model per query, reportedly cutting costs by around 60% for typical usage. The key trade-off is convenience versus control: Channels offers effortless setup while Telechat appeals to users with data privacy requirements or those preferring pay-per-use pricing.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Telechat lets you run Claude AI locally on Telegram, WhatsApp, and Slack

A developer has released Telechat, an open-source, self-hosted bot that connects Anthropic's Claude AI to Telegram, WhatsApp, Slack, and web chat from a single local process. Unlike Anthropic's own Claude Code Channels, Telechat routes no messages through third-party servers, relying instead on a personal API key to keep all data on the user's machine. The tool includes smart model routing that automatically selects the most cost-effective Claude model per query, which the developer claims cuts costs by around 60% compared to always using Sonnet. A standout feature called Desktop Bridge sends push notifications to a user's phone when Claude's coding agent requires tool-call approval, allowing remote approval via Telegram or WhatsApp. Telechat is available via npm and PyPI and is positioned as a privacy-focused, multi-platform alternative for users who need more control than Anthropic's first-party offering provides.

0 comments Read more at DEV Community