NetBay journal

AI Agents

30 practical AI Agents articles from Netbay Engineering — written for operators running Linux and Windows VPS in India.

30 articles in this topic

Fixture-Based Evals for Production Ops Agents

Score ops agents against frozen fixtures for SQL safety, command allowlists, and citation honesty so a model change cannot silently break production habits.

AI AgentsAugust 9, 20268 min read

When a Script Is Better Than an AI Agent

Choose a cron script over an AI agent when the steps are known, the failure is binary, and a Linux VPS should not spend tokens to restart nginx.

AI AgentsAugust 8, 20268 min read

Agent Identity, API Keys, and Secret Isolation

Give each agent its own Unix user, env file, and scoped API keys so a leaked worker key cannot drain the billing account or rewrite git history.

AI AgentsAugust 7, 20268 min read

Disk and CPU Watchdog Agent with Allowlists

Watch disk and CPU on a VPS and let the agent run only allowlisted fixes such as journal vacuum or nginx reload, logging every action it takes.

AI AgentsJuly 22, 20268 min read

Audit-Log Every Agent Tool Call in Production

Record who called which tool with which args, hash the result, and keep a replayable JSONL trail so incidents have evidence rather than guesses.

AI AgentsJuly 17, 20268 min read

A Server-Admin Agent That Runs Shell Safely

Build a server-admin agent that never gets a raw shell: allowlisted binaries, a cwd jail, timeouts, and no pipes, on a Linux VPS in Lucknow.

AI AgentsJuly 11, 20269 min read

Backup Verification Agent for Nightly Dumps

Prove last night dump is restorable: check age, size, archive headers, and a throwaway restore so backup success is a test, not a file timestamp.

AI AgentsJuly 3, 20268 min read

Retry, Timeout, Circuit Breakers for LLM APIs

Wrap every model HTTP call with timeouts, jittered retries, and a circuit breaker so one slow provider cannot stall every agent process on the box.

AI AgentsJuly 1, 20268 min read

Token and Cost Budgets for Long-Running Agents

Cap steps, tokens, and daily spend so a long-running ops agent on a Linux VPS cannot loop itself into an unbounded model-API bill overnight.

AI AgentsJune 13, 20268 min read

Structured JSON Output That Agents Can Parse

Force models to emit JSON Schema, then validate locally before any tool runs, so a stray fence or trailing sentence never crashes your agent loop.

AI AgentsJune 12, 20268 min read

Code-Review Agent That Posts PR Comments

Run a review agent on pull request diffs that comments on systemd, firewall, and secret leaks instead of style nits, then posts one GitHub review.

AI AgentsJune 11, 20268 min read

Evaluate AI Agents With Traces, Not Vibes

Stop judging ops agents by one lucky demo. Score JSONL traces against golden tasks so regressions show up before they restart the wrong systemd unit.

AI AgentsMay 30, 20269 min read