Does Together Link change my existing Claude Code or Codex settings?

No. Terminal agents get settings that last only for that session. Claude Desktop and ChatGPT Desktop use a separate profile, and `togetherlink claude-desktop off` or `togetherlink chatgpt off` switches back.

How much can I save, and how do I track spending?

Together Link cuts coding agent spend by 50-80% compared with running every session on Opus 5.5. Each session prints a cost receipt when you exit, and `togetherlink usage` shows your running total.

Together Link — Run open models in the coding harness you already use

Together Link is a free, open-source command-line tool from Together AI (currently in beta) that connects the coding agent you already use—such as Claude Code, Codex, or OpenCode—to open-weight models hosted on Together AI, including Kimi K3, GLM 5.3, MiniMax M3, and DeepSeek V4.1 Flash. It installs with a single command, keeps your existing agent logins, settings, and history intact, routes sessions automatically between cheap and capable models, and prints a token-and-dollar cost receipt at the end of every session.

Key Features

One-command setup

Install with `curl -fsSL https://link.together.ai/install | bash`, then run `togetherlink configure` with your Together AI API key and launch agents via per-agent commands like `togetherlink claude`. No config edits required.

Auto Router

Default routing mode that reads the first task in each session and sends quick work to a low-cost model (e.g., GLM 5.3) and harder problems to a more capable one. With an Anthropic API key in Claude Code or Claude Desktop, it routes between Opus 5.5 and GLM 5.3.

Per-session cost receipts

Every proxied session prints token totals and dollar totals when you exit, and `togetherlink usage` shows a running total covering the last seven days of spend.

Harness compatibility with preserved state

Works with Claude Code, Claude Desktop, Codex CLI, ChatGPT Desktop, OpenCode (v2+), and Pi (v0.80.8+). Your existing logins, settings files (~/.claude, ~/.codex, ~/.pi), history, projects, and other providers stay untouched.

Reversible profiles

Desktop apps (Claude Desktop, ChatGPT Desktop) run on a separate Together Link profile you can exit with commands like `togetherlink chatgpt off` or `togetherlink claude-desktop off`; terminal agents get settings that last only for the session.

Model choice from the agent's menu

Auto Router is the default, or you can pick a specific open model directly from your agent's model menu; Pi lets you cycle between open models mid-session.

How It Works

Install and configure

Run the one-line install command, then `togetherlink configure` to add your Together AI API key. Requires macOS or Linux and a supported coding agent.

Launch your agent through the Link

Run the agent's togetherlink command (e.g., `togetherlink claude`). For terminal agents, settings apply for that session only; desktop apps reopen on a separate Together Link profile with the Together model lineup added to the model menu.

Route and track

Set the model to Auto to let the router weigh each session's initial task, or pick a model manually. When you exit, the session prints its token and dollar receipt; `togetherlink usage` reports your recent spend.

Pros & Cons

Pros

Free and open-source tool; you pay only per-token model rates
Keeps your existing agent logins, settings, history, and other providers intact — fully reversible with a single command
Works across six harnesses: Claude Code, Claude Desktop, Codex, ChatGPT Desktop, OpenCode, and Pi
Auto Router balances cost and capability per session automatically
Transparent per-session cost receipts and a seven-day usage report
Access to frontier-quality open-weight models you control, at a fraction of closed API prices

Who It's For

Best for

Developers and teams already using Claude Code, Codex, OpenCode, or similar harnesses who want to cut token costs without switching tools
Cost-conscious teams running heavy daily coding-agent workloads who want open models and per-session spend visibility
Users who want frontier-quality open-weight models they control, with automatic routing between cheap and capable models

Not ideal for

Windows users
Teams that need harnesses outside the six supported agents
Users who want a flat-fee subscription rather than per-token billing

Use Cases

Moving everyday coding-agent work from closed frontier APIs to cheaper open models to cut agent spend by 50-80%

Running Claude Code, Codex, or OpenCode sessions on open-weight models like GLM 5.3, Kimi K3, MiniMax M3, or DeepSeek V4.1 Flash without changing tools

Automatic cost-aware routing so routine tasks hit a cheap model and hard problems escalate to a frontier one

Tracking per-session and weekly agent spend via receipts and the usage report

Pricing

Together Link itself is free and open source; you pay only Together AI's per-token rates for the models you use. Rates listed for the Link router: Auto Router $3.00 in / $15.00 out per 1M tokens; Kimi K3 $1.40 in / $4.40 out; GLM 5.3 $0.30 in / $1.20 out; MiniMax M3 $0.30 in / $1.20 out; DeepSeek V4.1 Flash also available.

Together Link (tool)

Free

Open-source tool; pay only for model usage at Together AI's per-token rates.

Auto Router

$3.00 per 1M input tokens / $15.00 per 1M output tokens

Default mode that routes every session between a low-cost model for quick work and a more capable one for hard problems.

Kimi K3

$1.40 in / $4.40 out per 1M tokens

Open frontier model selectable from your agent's model menu.

GLM 5.3

$0.30 in / $1.20 out per 1M tokens

Low-cost open model used by the Auto Router for quick work.

MiniMax M3

$0.30 in / $1.20 out per 1M tokens

Open frontier model selectable from your agent's model menu.

FAQ

What is Together Link?

A free, open-source tool that connects your coding agent (like Claude Code or Codex) to open models on Together AI, including Kimi K3, GLM 5.3, MiniMax M3, and DeepSeek V4.1 Flash. You pay only for model usage at Together AI's per-token rates.

Which coding agents does Together Link work with?

Claude Code, Claude Desktop, Codex, ChatGPT Desktop, OpenCode (version 2+), and Pi (version 0.80.8+). Each launches with its own command, such as `togetherlink claude`.

What do I need to get started?

A Together AI API key, macOS or Linux, and a supported coding agent. Install with `curl -fsSL https://link.together.ai/install | bash`, then run `togetherlink configure` to add your key.

How does the auto router choose a model?

Together Link

Key Features

One-command setup

Auto Router

Per-session cost receipts

Harness compatibility with preserved state

Reversible profiles

Model choice from the agent's menu

How It Works

Install and configure

Launch your agent through the Link

Route and track

Pros & Cons

Pros

Who It's For

Best for

Not ideal for

Use Cases

Pricing

Together Link (tool)

Auto Router

Kimi K3

GLM 5.3

MiniMax M3

Integrations

FAQ

Similar Products

Cons