togetherlink
Powered byTogether AI

Use Open source models
in Codex and Claude Code

Install once, then run Kimi K3 in Claude Code or Codex with short commands. TogetherLink injects Together settings for that run only, so your normal tool configs stay clean.

TogetherLinkper-run routing
Claude Codetclaude
ChatGPT Apptchatgpt
OpenCodetopencode
Codex CLItcodex
Pi Codetpi
Grok Buildtgrok
$curl -fsSL https://togetherlink.vercel.app/install.sh | sh
macOS & Linux - installs Bun if needed - keeps itself up to date
9
integrations
1
install command
GitHub
see the code
auto
updates in place
Beta

Claude Code

Claude Code routes through a local translation proxy - no edits to your real config. You keep your Claude Code subscription and login the whole time.

Beta

Codex CLI

Codex talks to Together through a local Responses-to-chat proxy. Settings are injected per run, with headless exec support for fast checks.

Alpha

DeepSeek Harness

DeepSeek Harness launches its web UI with a temporary Together provider patch. If DSH is missing, this explicit launch installs it first; normal DSH settings stay untouched.

100% supported

Grok Build

Grok Build launches directly on Together with a local metadata-only model catalog. Your native Grok config, built-ins, workflows, and sessions stay available.

Support100%
100% supported

Hermes Agent

Hermes Agent and Hermes Desktop launch with a temporary standalone Together provider. Sessions, skills, memories, plugins, and preferences stay available.

Support100%
100% supported

OpenCode

OpenCode launches with Together Kimi K3 wired in - config injected only for that launch. Close it and your OpenCode setup is exactly as it was, while sessions can still resume.

Support100%
100% supported

Pi Code

Pi Code launches with Pi's official Together provider, a temporary Pi config directory, and normal local session persistence.

Support100%
Beta

Prime Agent

Prime Agent launches with a generated Together provider extension and a runtime-only API key. Your native Prime settings, extensions, and sessions stay intact.

ChatGPT Desktop App

Alpha

Also works with the ChatGPT desktop app. Unlike the per-run CLI wrappers above, this persistently adds Together models beside native GPT models. Choose either provider from the same picker; restore removes the router and returns the exact prior config.

Configure

Adds Together models beside native GPT models in the same picker. The router stays active until you restore.

Restore

Removes the additive catalog and router, restoring the exact previous Codex config.

Heads upConfigure stays active until you restore. Backups live under ~/.togetherlink/backup/codex-app/.

Get started

$curl -fsSL https://togetherlink.vercel.app/install.sh | sh
1
Install with the one-liner above. It drops the binary at ~/.togetherlink/bin/ and adds togetherlink, tclaude, topencode, tcodex, tchatgpt, tdeepseek, tgrok, thermes, tpi, and tprime.
2
Run topencode, tclaude, tcodex, tdeepseek, tgrok, thermes, tpi, or tprime. Run thermes desktop for Hermes Desktop. For the ChatGPT desktop app run tchatgpt (alpha), and restore it with tchatgpt off. On first launch it asks once for your Together API key - press Enter to skip and add it later.
3
That's it. Your tool runs against Together models and stays up to date on its own. Change your mind? Just stop using it - no agent config was saved, so your subscriptions and your OpenCode/Claude Code/Codex CLI/DeepSeek Harness/Grok Build/Hermes/Pi Code/Prime Agent config are untouched.
Step-by-step

TogetherLink guides

View all →
Codex CLI

Run open models in Codex without replacing config

Route Codex through Together and switch between six current open models.

Grok Build

Launch Grok Build with GLM 5.2

Keep the Grok Build terminal harness while Together AI serves GLM 5.2.

ChatGPT Desktop

Add open models to the ChatGPT Desktop app

Configure the desktop app, understand the persistent change, and restore safely.