e-INFRA CZ LLM — Claude Code Integration¶
Country-specific: This guide is for Czech academic researchers only. Access requires Metacentrum / e-INFRA CZ membership. See https://metavo.metacentrum.cz for eligibility. This is not available for general international use.
What is e-INFRA CZ¶
e-INFRA CZ is the Czech national research e-infrastructure, providing computing, data storage, and — relevant here — free LLM API access to the Czech academic community. The LLM service is hosted at https://llm.ai.e-infra.cz and exposes an OpenAI-compatible API, which means Claude Code can use it as a drop-in replacement for Anthropic's API.
Who can access¶
Czech academic researchers with an active Metacentrum / e-INFRA CZ membership. Membership is free for researchers affiliated with Czech universities and research institutions. Register or check your status at https://metavo.metacentrum.cz.
Available models¶
| Model | Category | Notes |
|---|---|---|
qwen3.5-122b |
General / best overall | Default recommended |
qwen3-coder-next |
Coding | Latest coding model |
deepseek-v3.2 |
Coding + reasoning | Good all-rounder |
deepseek-v3.2-thinking |
Reasoning / thinking | For complex reasoning |
kimi-k2.5 |
Agentic / tool use | Best for tool-calling workflows |
mistral-small-4 |
Fast / small | Quick tasks |
qwen3-coder-30b |
Fast / coding | Lighter coding model |
Mode A — Direct connection (simplest)¶
Set two environment variables and launch Claude Code:
⚠️ Limitation: Claude Code sends claude-* model names in all requests. The e-INFRA API accepts these in direct mode, but you cannot select a specific e-INFRA model — the API uses its default routing. If you need a specific model (e.g. kimi-k2.5 for agentic workflows), use Mode B.
Note on model name in responses: In direct mode the e-INFRA API may return a response with a non-Claude
modelfield (e.g."model": "qwen3.5-122b"). Claude Code validates this field and may error with "unexpected model". If this happens, switch to Mode B (proxy) — the proxy automatically restores the originalclaude-*model name in every response so Claude Code is satisfied.
Mode B — Proxy (model selection)¶
The proxy script intercepts requests from Claude Code, replaces the model name with your chosen e-INFRA model, and forwards the request to the e-INFRA API. This lets you select any model from the table above.
Step-by-step terminal workflow¶
Step 1 — Copy the proxy script
Copy skills/setup/scripts/proxy.mjs from the neuroflow plugin to a convenient local folder (e.g. ~/tools/einfra/).
Step 2 — Add your API key
Open the copied proxy.mjs and replace <YOUR_API_KEY> with your actual e-INFRA API key.
Step 3 — Terminal 1: start the proxy
node proxy.mjs # default: qwen3.5-122b, port 3456
node proxy.mjs kimi-k2.5 # use kimi-k2.5
node proxy.mjs deepseek-v3.2 8080 # custom model + port
The proxy prints a confirmation:
Step 4 — Close Claude Code
If Claude Code is currently running, close it completely before proceeding. Environment variables are read at startup — Claude Code will not pick them up if it is already open.
Step 5 — Terminal 2: start Claude Code pointing at the proxy
Note: ANTHROPIC_API_KEY=any is a placeholder — Claude Code requires a non-empty value, but the proxy handles the real API key. The value itself does not matter.
Step 6 — Verify
Claude Code is now routing all requests through the proxy → e-INFRA. The proxy terminal will print each request as it arrives.
For persistence (shell profile)¶
Direct mode — add to ~/.zshrc or ~/.bashrc (macOS / Linux):
Then restart your terminal and reopen Claude Code — no proxy needed.
Proxy mode — start the proxy in a background terminal each session, or add it to a launch script. The ANTHROPIC_BASE_URL=http://localhost:3456 pointing to the proxy must be set in the shell where you open Claude Code.
Windows users¶
Node.js runs proxy.mjs on Windows without changes. The only difference is how you set environment variables.
PowerShell (recommended):
# Direct mode — set for this terminal session, then launch Claude Code
$env:ANTHROPIC_BASE_URL = "https://llm.ai.e-infra.cz/v1"
$env:ANTHROPIC_API_KEY = "<YOUR_API_KEY>"
claude
# Proxy mode — open two PowerShell windows
# Window 1: start the proxy
node proxy.mjs kimi-k2.5
# Window 2: launch Claude Code
$env:ANTHROPIC_BASE_URL = "http://localhost:3456"
$env:ANTHROPIC_API_KEY = "any"
claude
For persistence on Windows: add to your PowerShell profile (notepad $PROFILE):
cmd.exe (legacy):
⚠️ Parallelization note¶
Custom API providers — including e-INFRA — can freeze or time out when Claude Code fires many concurrent tool calls. neuroflow's scholar agent was specifically updated to run searches sequentially (PubMed → bioRxiv → fallbacks, one at a time) and download papers in batches of 2 for this reason.
If you experience freezing or silent hangs with other neuroflow workflows while using e-INFRA, reduce concurrency manually: avoid commands that trigger many simultaneous file reads or API calls, and prefer sequential workflows over parallel ones.
Saving to neuroflow setup¶
The /neuroflow:setup Step 5 saves the following to .neuroflow/integrations.json under the custom_llm key:
"custom_llm": {
"provider": "einfra",
"base_url": "https://llm.ai.e-infra.cz/v1",
"api_key": "<YOUR_API_KEY>",
"model": "qwen3.5-122b",
"proxy_port": 3456
}
api_keyis stored locally and gitignored — it is never synced anywhere.- Non-secret fields (
provider,base_url,model,proxy_port) can optionally be written to your linked flowie private GitHub repo for cross-machine sync. Run/neuroflow:setupStep 5 to configure this.