TierMux

Stack free. Route smart. Ship faster.

TierMux is a free, open-source AI coding assistant for VS Code. Instead of locking you into a single model or running up an API bill, TierMux automatically routes each message to the best available free model across 22+ providers — and silently switches to another one when a provider is slow, rate-limited, or down.

The name: Tier (free provider tiers) + Mux (a multiplexer that switches between them).

What Problem Does TierMux Solve?

Most AI coding tools force a trade-off:

• Pay per token → bills grow fast, especially with agentic tasks that make dozens of requests.
• Pick one model → you're stuck with its rate limits, downtime, and blind spots.
• Manage keys manually → juggling 10 dashboards is not how you want to spend your time.

TierMux eliminates all three. It runs entirely on the free tiers of 22+ providers, automatically picks the right model for each task, and fails over silently when any one provider has issues. You just type — TierMux handles the rest.

Supported Platforms

TierMux connects to 22+ AI providers out of the box. Each one is pre-configured — you just add a key (or use it keyless where supported):

⚡ Keyless means you can use the provider with zero configuration — no account needed.

You can also add your own OpenAI-compatible endpoint (vLLM, LiteLLM, Azure OpenAI, Cloudflare AI Gateway, etc.).

How TierMux Learns — What It Understands About Your Work

TierMux isn't just a dumb proxy. It actively learns from your workflow:

🔀 Smart Routing — Right Model, Right Task

Every message is classified before it's sent. A quick "what does this function do?" goes to a fast, lightweight model. A "refactor this entire auth system" goes to a deeper reasoning model. You don't need to think about this — TierMux decides automatically when you're on Auto mode.

🔁 Automatic Failover — Never Gets Stuck

If a provider rate-limits you, times out, or returns an empty/refused response, TierMux silently tries the next best option in your priority list. Background agents keep running. You never see an error unless every provider in the chain fails.

🧠 Learns Your Preferences

• Give any reply a 👍 or 👎 — TierMux remembers which models actually work well for your codebase and adjusts future routing.
• Picks up on your coding style (indentation, quote style, semicolons) directly from files you edit and instructs the model to match it — no manual config needed.

📊 Tracks Your Savings

Every token request is tracked. The footer shows live token usage and an estimate of how much money you've saved compared to paying for a commercial API. Across a day of agentic coding, this can easily reach $5–$50+.

What TierMux Can Do For You

Three Modes

Leave everything on Auto and TierMux reads your message to pick the right mode and the right model automatically.

Agent Capabilities

• Reads your project first — greps files, checks types, reads diagnostics — so it's not guessing about your codebase.
• File editing with diffs — every change shows up as a diff you review and approve before it's applied.
• Terminal commands — runs commands with configurable confirmation levels (always / auto-approve safe / fully autonomous).
• Checkpoint system — saves a snapshot every turn so you can undo back to before any message.
• Fallback on bad answers — if a model returns empty output, refuses, or loops, TierMux automatically retries with a smarter model.
• Parallel agents — multiple chat tabs can each run their own agent simultaneously (up to your configured limit).

Editor Integration

• Right-click menu → Explain / Fix / Refactor / Generate Tests / Generate Docs on any selected code.
• Fix with AI on red squiggly errors directly from the Problems panel.
• Inline chat → Ctrl/Cmd+I anywhere in the editor for instant in-place edits.
• Git commit messages — TierMux writes them for you based on your staged diff.
• Optional inline autocomplete — off by default to preserve free-tier quota, but available to enable.

MCP Server Support

Connect Model Context Protocol (MCP) servers to give the agent additional tools — databases, APIs, file systems, custom scripts, and more.

Getting Started

Step 1 — Open TierMux

Click the TierMux icon in the VS Code Activity Bar (sidebar).

Step 2 — Configure Providers (UI-Based)

Click ⚙ Manage Models & Keys inside the TierMux panel.

The configuration UI lets you:

• Enable / disable individual providers with a toggle
• Add API keys for each provider in a secure input field
• Set priority order — drag providers up/down to control which ones TierMux tries first
• Add custom endpoints — paste any OpenAI-compatible base URL
• Set custom headers for enterprise endpoints that require them

No settings.json editing required. Everything is point-and-click inside the extension panel.

Step 3 — Start Coding

Leave everything on Auto and start typing. If you have no keys yet, start with a keyless provider like OVH, Pollinations, or Kilo — no account required.

Settings

All settings live inside the TierMux panel — no need to open VS Code's settings.json.

• Providers & Models → ⚙ Manage Models & Keys tab
• Agent behavior, context, memory, timeouts → Others tab in the same panel

🔒 Your Privacy & Credentials

Your API keys never leave your machine.

• Keys are stored in VS Code's encrypted secret storage — the same secure vault VS Code uses for Git credentials. They are never written to any config file, never logged, and never synced to the cloud.
• No TierMux backend server exists. There is no middleman. Every request goes directly from your VS Code instance to the model provider's API.
• Feedback and usage stats are stored locally only — in your VS Code extension storage folder on your own machine.
• Nothing you type, paste, or ask is ever sent to TierMux — only to the provider you've selected (or that TierMux auto-selected for that message).

You are always in control of which providers are active, which models are used, and how your keys are stored. TierMux is a routing layer — not a service.

About TierMux

TierMux was built to answer a simple question: why should writing code with AI cost money when so many excellent models are free?

The free tiers of modern LLM providers — Groq, Cerebras, Google AI Studio, OpenRouter, Mistral, NVIDIA NIM, and many more — are genuinely powerful. The problem is that no single free tier is reliable enough on its own: rate limits hit, providers go down, and different models are better at different things.

TierMux solves this by treating all those free tiers as a single, unified, self-healing pool. It routes intelligently, fails over automatically, and learns what works best for your specific codebase and style. The result is a coding assistant that costs nothing, improves over time, and never leaves you staring at a rate-limit error.

License

MIT — free to use, modify, and distribute.