FAQ
Questions, answered.
Models, datasets, licensing, the gateway and privacy. If something is still unclear, the contact form reaches the founder directly.
What models does Tantra AI Labs release?
Two open-weight 7B models: AstraGPT-7B (base, reasoning) and AstraGPTCoder-7B (coding sibling). Both are Apache-2.0 and published on Hugging Face under the tantra-ai-labs organisation.
Are the models free to use?
Yes. AstraGPT-7B and AstraGPTCoder-7B are released under Apache-2.0, which permits commercial use, modification and redistribution under the license terms. The license on each Hugging Face repository is the authoritative document.
How were the models trained?
A 7B decoder-only transformer with a custom BPE tokenizer, grouped-query attention (28 query / 4 KV heads), RoPE at θ=1M and a SwiGLU feed-forward. Reasoning fine-tuned with LoRA on 193,841 instruction-reasoning-answer triples using 2× RTX 4090 GPUs. Full details are on the AstraGPT-7B card.
What is in the reasoning dataset?
193,841 instruction-reasoning-answer triples with explicit Attempts Blocks traces: exploration, dead ends, self-correction and commitment, followed by the answer. Apache-2.0, available on the datasets page.
Can I use the datasets commercially?
The AstraGPT Reasoning Dataset is Apache-2.0 and the High-Reasoning Dataset v1 is MIT. Both permit commercial use under their terms; check the dataset card for attribution requirements. See datasets.
How do I get API access?
ModelRouter is the production gateway: one OpenAI-compatible endpoint with 40+ models, streaming and fallbacks. Create a key on modelrouter.in and point your client at
https://api.modelrouter.in/v1.Which models can I call through the gateway?
The gateway routes 40+ models across multiple providers and self-hosted vLLM deployments. The exact catalogue depends on your key's allow-list and is visible in the ModelRouter dashboard.
Do you take consulting or security work?
Yes — model training, inference infrastructure and offensive-security reviews are handled directly by the founder. Use the contact form with a short description of the work.
How is my contact information handled?
Contact-form data (name, email, optional phone, subject, message, IP and user-agent) is stored on a server the lab operates and used only to reply and prevent abuse. No analytics, no advertising trackers, no selling data. See the privacy policy.
Is the website open source?
The models and datasets are open under their licenses. The website source and the gateway implementation are not published as open source.
Still stuck?
Send a message — it reaches the founder, not a queue.