Credit
The Independent 100 list was curated by dan (@irl_danB) and published on 2026-09-24 at independent.prose.md (announcement). The ranks and section names are his, and number 100 is his open wild-card slot. This study is ours and is not endorsed by him or by anyone on the list.
What they publicly do
[Verified] Works on Moondream, a compact open vision-language model (moondream.ai). His site says he spent nine years on distributed systems at AWS before Moondream (vikhyat.net). [Verified] Moondream's code lives under the m87-labs GitHub org (github.com/m87-labs/moondream); the 'Elbows out' post on his site announces Moondream's seed round.
- X profile: x.com/vikhyatk
- GitHub: github.com/vikhyat
- Also: vikhyat.net
- Also: moondream.ai
- Also: github.com/m87-labs
Public work
- moondream repo Apache-2.0 last activity 2026-04-20 · 10,062 GitHub stars on 2026-09-24
Tiny vision language model: code, client and examples. vikhyat/moondream redirects here.
- Moondream 3.1 (9B total, 2B active MoE) model Moondream Model License 1.0 (source-available, not OSI) last activity 2026-07-08
VLM with native query, detect, point and caption skills that return structured output. Runs locally through Photon on NVIDIA GPUs and Apple Silicon.
- Parakeet Redux model CC-BY-4.0 last activity 2026-09-23
1.58-bit (ternary) parakeet-tdt-0.6b-v3 ASR in 178 MB. The model card reports 113x real time on 8 x86 cores and runs on NEON, Metal and AVX-512 VNNI through Photon. English WER 6.55 vs 6.26 for the original.
- Parakeet Ultra model CC-BY-4.0 last activity 2026-09-23
Full-precision sibling of Redux, trained further, for GPUs.
- kestrel repo no license file last activity 2026-09-24 · 100 GitHub stars on 2026-09-24
The inference engine behind Photon: async micro-batching, paged KV cache, radix prefix cache, LoRA, and CUDA and Metal kernels.
- moondream2 / md3p-int4 (MLX) model Apache-2.0 last activity 2025-12-19
Earlier Apache-2.0 weights, including an MLX int4 build of Moondream 3 preview.
Licenses of GitHub repositories are from the authenticated GitHub license API on 2026-09-25, one row per repository in evidence/licenses.jsonl (“no license file” means the API found none; “license file present, not recognised by GitHub” means GitHub could not classify it). Other licenses are as the source states. Check the license file before reusing anything.
Relevance to our work
5 of 5 lenses: local-models, content, search
[Verified] Moondream ships locally runnable vision and speech models with stated licenses and Apple Silicon support. Parakeet Redux (CC-BY-4.0, 2026-09-18) is a direct benchmark candidate against local ASR such as franken_whisper on Apple Silicon. Moondream 3.1's detect, point and caption skills fit media asset tagging and screenshot QA for agents. [Verified] The license allows self-hosted and SaaS-component use.
Scored 0 to 5 against the study’s relevance rubric: 5 means public, openly licensed work squarely in our lanes that we could study or adopt now, and 1 means little connection to our work. The score is one reviewer’s judgement of how much this public work bears on our own work. It is not a ranking of people and says nothing about the person or the quality of their work.
What we would study next
- Benchmark Parakeet Redux against franken_whisper and whisper on narration and meeting audio (huggingface.co/moondream/parakeet-redux). A 178 MB ternary ASR that claims near-original English WER on CPU and Metal: a possible faster local transcription path. Effort: medium.
- Moondream 3.1 detect/point/caption as a screenshot and video-frame labeller for agents (moondream.ai/blog/moondream-3-1-beyond-benchmarks). Structured visual output could back visual QA and media asset manifests without a frontier API. Effort: medium.
- Kestrel architecture (micro-batching, radix prefix cache, Metal kernels) (github.com/m87-labs/kestrel). A reference design for a local multimodal serving layer; compare with MLX-based serving. Effort: medium.
Before reusing anything: The moondream code repo is Apache-2.0. Moondream 3.x weights use the Moondream Model License 1.0: source-available, not OSI. It allows commercial, SaaS-component, fine-tuning and self-hosted use, but a general-purpose hosted Moondream inference or fine-tuning service needs a commercial license (Section 3). A domain-specific deployment is allowed; a generic Moondream endpoint is not. The license also requires Delaware venue and has a patent-termination clause. Parakeet weights are CC-BY-4.0 (attribution required). kestrel has no license file per the GitHub license API (2026-09-25) [Inference: all rights reserved]. The Photon runtime's own license was not checked.
Sources
- vikhyat.net
- api.github.com/repos/m87-labs/moondream
- huggingface.co/moondream/moondream3.1-9B-A2B
- moondream.ai/licenses/model/1.0
- huggingface.co/moondream/parakeet-redux
- huggingface.co/api/models?author=moondream&sort=lastModified&limit=10
- github.com/m87-labs/kestrel
Overall confidence in this record: High. Sources were read on 2026-09-24 and every link was checked on 2026-09-25. The full record is line 54 of people.jsonl.
Limits
Not verified: The license of the Photon runtime and the moondream pip package. The Parakeet Redux speed and WER numbers are the vendor's own and have not been reproduced. vik's exact title at M87 Labs (his site says only that he spends most of his time on Moondream).
This record comes from a first-pass survey of public material by an AI agent session on 2026-09-24, edited for publication on 2026-09-25. Nothing was installed or run for it. Roles, stars and licenses change; see the study’s limits.
Request a correction
Is something about this record wrong or out of date, or would you rather not be covered? Open a correction (a GitHub issue form with this page’s address filled in). A request to remove a record is honoured without argument; accepted corrections are recorded in the changelog with the date.