LIVE
+++ Ox Alpha Unmasked: GLM-5.3-Flash Lifts Zhipu Shares +++ OpenGrok opens Grok Bot to third-party models +++ Sepia targets the structure of AI-written text +++ Microduck: Hugging Face's $399 Open-Source Robot Duck +++ Gemini Omni 1.1 Flash adds a $0.03 draft mode +++ Who gains from OpenAI's executive exodus? Brockman ++++++ Ox Alpha Unmasked: GLM-5.3-Flash Lifts Zhipu Shares +++ OpenGrok opens Grok Bot to third-party models +++ Sepia targets the structure of AI-written text +++ Microduck: Hugging Face's $399 Open-Source Robot Duck +++ Gemini Omni 1.1 Flash adds a $0.03 draft mode +++ Who gains from OpenAI's executive exodus? Brockman +++
All news ›
AI IN LIFE AI IN LIFENEWS
DAILY
DE EN
MODELS

OpenGrok opens Grok Bot to third-party models

An MIT-licensed community project swaps the model behind Grok Bot. Six providers are marked working, three still await a wire capture.

OpenGrok opens Grok Bot to third-party models

A single connector fanning out into nine identical ports — an image for one bot rerouted to several model providers; the illustration is AI-generated.

OpenGrok is an MIT-licensed community project that reroutes Grok Bot to outside providers — GLM, Claude, Gemini, DeepSeek or a local llama.cpp — through a model picker, and says it has no affiliation with xAI.

At a glance

  • MIT license; the OnlyTerp/opengrok repository shows 279 stars and 29 forks as of August 29, 2026.
  • Six providers are marked working: Grok, GLM, Claude, Gemini, DeepSeek and local llama.cpp.
  • Three providers still await a capture per the status table: OpenRouter, Groq and Mistral.
  • Setup is git clone plus python setup.py; the script says it asks three questions.
  • tools/doctor.py baselines the machine at setup and reports what moved after an update.

OpenGrok is an MIT-licensed community project that puts a model picker in front of Grok Bot and sends requests to outside providers instead: GLM, Claude, Gemini, DeepSeek, or llama.cpp running locally. The project describes itself as unaffiliated with xAI. Everything below comes from the repository page — no second newsroom has covered this, and we say so rather than papering over it.

What the tool actually does

The center of it is a dropdown: pick a model per agent, test it, save it. The test fires a live request at the provider rather than merely validating a config file. Behind that sit two routing files, provider-maps.cjs and provider-maps-hop.cjs, which describe what a request to each provider has to look like.

Installation is three steps: clone the repository, change into the directory, run python setup.py. The script says it detects existing installations, asks three questions, then opens the picker. The page does not spell out what those three questions are.

Which providers work, and which do not yet

The project keeps a status table and, to its credit, draws a hard line between evidenced and unproven. Six entries are marked working; three are not:

ProviderStatus per the project
Grok (xAI)working
GLM (Zhipu)working
Claudeworking
Geminiworking
DeepSeekworking
llama.cpp (local)working
OpenRoutercapture pending
Groqcapture pending
Mistralcapture pending

That split is the genuinely interesting part. The project's stated rule is that no provider map ships without a real capture of the request behind it. Captures are produced by tools/wire-probe.py and stored under wire-captures/; the page cites a capture for GLM 5.3 Flash as its example.

A doctor for other people's updates

The second tool is tools/doctor.py. It takes a baseline of the machine during setup, then watches files, services and caches. When a new Grok Bot release shifts something underneath, the tool is supposed to name what moved. Provider maps are meant to hot-reload without a restart.

Alongside those sit tools/qa.py for a repository self-check including a scan for leaked credentials, plus tools/apply-box-patch.py and tools/file-relay.py for cloud hosts. The project flags that cloud case as a limitation of its own: stock Grok Bot cloud hosts do not read model-bindings.json on their own, so an extra patching step is required.

Size and maturity

On August 29, 2026 the repository page shows 279 stars, 29 forks, 14 commits on the main branch, one open issue and no open pull requests. Listed topics include ai-agents, grok, llm, model-router and openai-compatible. A badge claims Windows, macOS and Linux; no minimum Python version appears on the page.

The project's tagline — "Not farming you, arming you" — is aimed at credential handling: API keys are meant to stay on your own machine.

What one source cannot settle

These points stay open, and we are not rounding them up into claims. First, the promise that keys never leave your machine is the project talking about itself; we have no independent audit of the code. Second, we ran neither setup.py nor wire-probe.py, so we cannot say the stored captures reproduce.

Third, the lack of xAI affiliation is a self-declaration, not a statement from xAI. Fourth, we have nothing from any of the companies involved on whether this kind of rewiring sits within Grok Bot's terms of service. Fifth, no second independent outlet has reported on the project at all.

◈ AI-GENERATED REPORT · SOURCES LINKED

FAQ

What is OpenGrok?

An MIT-licensed community project that adds a model picker to Grok Bot and forwards requests to other providers. It does not come from xAI.

Which models does OpenGrok support in Grok Bot?

Marked working: Grok, GLM, Claude, Gemini, DeepSeek and a local llama.cpp. OpenRouter, Groq and Mistral are still listed as awaiting a capture.

Is OpenGrok an official xAI tool?

No. The project states it is not affiliated with xAI. We have no confirmation or denial from xAI itself.