baseten

Portal hopping with Stable Diffusion 👾

Mantis - unified perception-reasoning-action agent for computer use

Open, independent benchmark on LLM inference providers: same GLM 5.3 Flash, 600 paired requests each. Latency, tokens per second and task success. Baseten, DeepInfra, Fireworks AI, Modal, Nebius, Novita AI, Parasail, Telnyx, Together AI, Z.AI. Python runner and public data.

Span-level prompt-injection guard for AI agent tool traffic: strip the attack, keep the data. Built for Meta Global AI Hackathon 2026 — fine-tuned open-weights span model, committed benchmarks vs Llama Prompt Guard 2.

The open control plane for multi-model AI

Precision load intelligence for disaster debris removal. Video-verified volumetric tickets & defensible FEMA audit records powered by Meta SAM 3.1 (spatial segmentation), Muse Spark 1.3 (multimodal OCR & fill estimation), and Muse Voice (audio radio calls). Built for the Meta Global AI Hackathon 2026.

AI system integrator that modernizes legacy COBOL to Java and SQL, built on Meta Muse Spark

AI pathology pre-reads for Kenyan pathologists, and outbreak alerts from confirmed results. A Flutter app for patients, doctors and pathologists powered by Firebase, Meta, Cerebral Valley, Baseten, Modal, and Fireworks; every AI draft is approved by a qualified person before it reaches anyone. Piloting across 1,447 Nairobi facilities.

Agent Beans: an always-on AI growth team for Meta ads, for small businesses and indie developers. Built on Meta's Muse models.

Life: an autonomous, self-improving agent. It plans your day, sets your alarms and limits, steps in when you're stuck scrolling, and learns every night what works on you. Android + Meta Muse Spark, Muse Spark Open, Muse Image & Voice Transcribe.

Hearth: one calm place for your day, health and AI agents. Built for Baseten and Modal. Built with Muse (Meta).

One-shot GPU harness for traffic-grounded teacher, student, and route evaluation

Agora: decks, evidence and setup guide for Teleion Fabric, infrastructure for AI agents, and Eranos.

Noa, the relationship AI that remembers the little things and helps you show up. Built with Meta Muse Spark, Muse Spark Open on Baseten, and Muse Code for the Meta AI Hackathon.

Kosmos: VLMs for world-model robot-policy evaluation.

Supervisory layer for AI coding agents: parks approval requests, keeps independent work moving, detects repeated failures, gets a Meta AI diagnosis, and verifies recovery with real test results. Go + SQLite.

Synthetic event demo explaining deterministic quote-gate audits with Meta Muse Spark Open on Baseten (Meta Global AI Developer Hackathon)

Agent swarm that bisects vLLM on a real A100, names the commit that slowed your Llama endpoint, and proves it with statistics. Open weights end to end (Llama-3.2-1B + muse-spark-open).

ClipCraft Studio is an AI-powered short-form video content generation studio that transforms ideas into researched stories, reviewed scripts, scene-by-scene video prompts, character direction, thumbnails, and platform-ready metadata.

Permissions-aware RAG reference implementation over Canadian federal public documents. Docker Model Runner, pgvector, Compose, Mattermost and MCP front doors, inference benchmarked locally and on Baseten.