Zed + JoinGonka Gateway — cheap inference in your editor

Zed is a fast open editor with an Agent Panel, and it supports any OpenAI-compatible provider: just specify your api_url and a list of models. This means Zed's agent features can run on Gonka decentralized network models via JoinGonka Gateway — at live network prices, hundreds of times lower than vendor card prices, with starting 3M free tokens without needing a credit card.

Step 1: Get a JoinGonka key

Register at the gateway and create a jg-… key. The current list of network models is always available via GET https://gate.joingonka.ai/v1/models; today these are MiniMaxAI/MiniMax-M2.7, deepseek-ai/DeepSeek-V4-Flash-0731, and zai-org/GLM-5.3-Flash.

Step 2: Add a provider via UI or settings.json

Via installer (fastest):

npx @joingonka/setup --tool zed

This will append the provider and model catalog directly to settings.json, preserving your comments and other editor settings. The key is not placed there — Zed does not read it from there (see below).

Via UI: Open agent settings (command agent: open settings) → in the "LLM Providers" section, click Add Provider → fill in the fields: provider name, api_url = https://gate.joingonka.ai/v1, the jg-… key, and the models. The key goes into the system keychain, not the text config.

Via settings.json (command zed: open settings) — the language_models.openai_compatible block:

`api_url`	`https://gate.joingonka.ai/v1`
`available_models`	array of models; for each — `name`, `max_tokens` (this is the CONTEXT, not the output limit) and `max_output_tokens`

Correct limits for network models: DeepSeek V4 Flash — context 380000, output 32768; MiniMax M2.7 — context 200000, output 8192; GLM-5.3 Flash — context 390000, output 8192 (reasoning model: part of the output is used for reasoning, do not lower the limit).

Key. It is not written to settings.json — Zed documentation explicitly forbids this, and there is no field for it in the schema. Provide the key in one of two ways: via the JOINGONKA_API_KEY environment variable (Zed derives the name from the provider ID) or insert it in the provider panel — then it will be stored in the keychain.

Step 3: Select a model and start working

Added models appear in the Agent Panel model switcher—the model is selected per conversation. All three network models support tool calling, so Zed's agent functions (code edits, multi-step tasks) work as expected. For large repositories and long sessions, use DeepSeek V4 Flash: one of the longest contexts in the network (380K) and output up to 32768 tokens. Formally, the longest context (390K) belongs to GLM-5.3 Flash, but part of its response budget goes to reasoning.

What works on a custom endpoint and what doesn't

Agent Panel, chat, and inline assist work with your provider. Edit predictions (Zeta autocomplete) remain on Zed's own models — third-party APIs cannot be connected there; this is a limitation of the editor, not the gateway. Vision tasks do not work through the gateway: network models are text-based.

How much does it cost

Inference is billed at live Gonka network prices — see current values in the gateway price list. A uniform rate for all network models and starting 3M free tokens allow you to test the Zed agent on real tasks before your first top-up.

Zed connects to JoinGonka Gateway in a couple of minutes: Add Provider in agent settings (or settings.json block) → api_url gate.joingonka.ai/v1 → key jg- in keychain → network models in the Agent Panel selector. Edit predictions remain on Zed's models; everything else runs on cheap Gonka inference.

← GitHub Copilot BYOK — own models instead of quotas Pi + JoinGonka Gateway — terminal agent on cheap inference →

Want to learn more?

Explore other sections or start earning GNK right now.

Get free tokens →