Agastya AGI v01 Native Brain Evidence Release R1

Agastya AGI v01 is a sovereign, from-scratch native foundation-model program developed in India by Rabbit Industries Pvt Ltd under the Bramha Medha Native INDIA family. Agastya Native Brain Evidence Release R1 publishes the cryptographically anchored provenance of the canonical Agastya Native Brain lineage and is intended to make the project externally citable, reproducible at the evidence layer, and suitable for independent technical verification. The canonical development lineage is: Agastya corpus → Agastya tokenizer → Agastya architecture → RANDOM INITIALIZATION → Agastya pretraining → Agastya post-training → Agastya Native Brain The evidence package records and publishes: Fresh random initialization at step 0 — PASS Teacher-weight ancestry — ZERO Third-party weight ancestry — FALSE Third-party tokenizer artifact import — FALSE Unique parameter count — 1,264,715,776 Architecture class — dense native model Canonical context length — 4,096 tokens Native tokenizer vocabulary — 98,320 entries Tokenizer, architecture, initialization and checkpoint provenance — SHA-256 anchored Checkpoint validation — PASS Clean-P0 zero-third-party acceptance — PASS The release includes a technical evidence report, machine-readable weight-lineage and tokenizer-provenance manifests, checkpoint-validation evidence, Clean-P0 acceptance evidence, source-artifact hashes, package checksums, citation metadata, and an independent-verification protocol. The canonical Agastya Native Brain does not inherit, merge, distill, or initialize from third-party model weights. Reference models may be used separately for evaluation and comparative research, but they are not part of the canonical Agastya weight ancestry. This public evidence release intentionally does not redistribute proprietary raw model weights, the raw training corpus, the complete tokenizer vocabulary/merges artifact, private architecture source code, or proprietary training source code. These artifacts are retained by Rabbit Industries and can be made available under controlled conditions to qualified independent evaluators for provenance verification. Independent third-party verification of the complete lineage is the next external validation gate. Same-suite public benchmark qualification against contemporary frontier systems will follow using frozen model versions, evaluation harnesses, prompts, scoring procedures, tool permissions, and reproducible environments. No independently verified global performance rank is claimed in this R1 release. Rabbit Industries is pursuing independent verification, DOI-backed preservation, public model documentation, frozen same-suite benchmark evaluation, and external leaderboard acceptance so that future capability claims are supported by directly comparable evidence. India positioning: Agastya is developed in India by Rabbit Industries Pvt Ltd as a from-scratch sovereign native foundation-model program with publicly released cryptographic provenance evidence. Rabbit Industries is pursuing independent prior-art and technical verification for a narrower India-first distinction based on independently verified native lineage, zero third-party weight ancestry, and independently verified tokenizer provenance. Global objective: Agastya is designed and being evaluated for global frontier competitiveness while maintaining sovereign model provenance and native development principles. Reserved DOI:10.5281/zenodo.23054425 Canonical public evidence:https://zippri.com/agastya-native-brain-evidence/ Public benchmark profile:https://zippri.com/agastya-benchmark-profile.html Competition reference:https://zippri.com/agastya-competition.html Publisher: Rabbit Industries Pvt LtdModel: Agastya AGI v01Family: Bramha Medha Native INDIAEvidence release: R1