Agentic
Lets you keep using the Claude Code interface you already know while quietly routing requests to cheaper or different models, with a hard budget cap so you never get an unpleasant surprise bill.
🔗 Visit Agentic
Description
Claude Code is a specific interface tied to Anthropic's models by default, which is a problem if you want to mix in a cheaper model for simple tasks or a different provider entirely without relearning a new tool. Agentic sits between you and Claude Code, keeping the same interface while routing each request to Anthropic, OpenAI, xAI, Ollama, vLLM, OpenRouter, DeepSeek, or Groq depending on what you configure.
It's a static Go binary with no daemon required, and it includes real budget management: daily, weekly, and monthly spending caps, token tracking with pricing data on every request, and detailed cost breakdowns by model, profile, or session. An LLM-powered classifier automatically sorts tasks into deep, standard, or light tiers so trivial requests don't burn an expensive model's budget. It supports leader election for sharing across multiple instances and hot-reload configuration, plus a `agentic cost` CLI command for spending reports. It's fully open source under the MIT license.
💬 Our review
The short version: if you've ever gotten a surprise API bill from an agentic coding session left running unattended, the budget caps here solve a real, specific pain point that the underlying tools (Claude Code, OpenRouter, etc.) don't solve on their own.
Against using Claude Code directly with a single provider, Agentic's value is flexibility (multi-provider routing) plus hard spending limits — neither of which the base tool offers. Against a general LLM gateway like OpenRouter used standalone, Agentic's differentiator is staying inside the Claude Code interface rather than requiring you to switch tools, and its automatic task-tier classification tries to route cheap tasks to cheap models without manual intervention. Being free, open source, and a single static binary means there's essentially no cost to trying it if you're already a Claude Code user worried about spend.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Open source, licence MIT
Pros
Plafonds de dépense quotidiens/hebdo/mensuels intégrés
Routage multi-fournisseurs tout en gardant l'interface Claude Code
Binaire Go statique unique, sans daemon requis
Cons
Réservé aux utilisateurs déjà habitués à Claude Code
Projet indépendant récent, pas de support commercial
Classificateur de tâches par IA ajoute une dépendance supplémentaire