Past 3 days: Dev tools for AI agents lead 12 launches; Rozana's $31.6M Series B dominates capital flow. Open-source tooling and infrastructure abstraction becoming the race-to-scale pattern.

GL OO CC AO M

Five launches in 72 hours (Git-lanes, OculOS, Context-compact, Agent Office, Max) signal builders racing to abstract OS/desktop APIs for agentic automation. OculOS converts desktop accessibility trees to JSON; Context-compact optimizes token use in agent prompts. No VC backing yet, but momentum suggests early PMF stage before Series A wave. Gap: None of these solve real-time handoff between autonomous agents and human overrides—all are agent-first, agent-only.

NN IO M S

Four gaming launches (NeoNetrek, I open, Moongate, Swarm) clustered on day 5 of the window. All open-source, all GitHub-first. No funding signal. Likely community-driven preservation or indie game engine iteration, not venture-scale. Gap: Gaming as a vertical has stalled capital flow despite AI-generated content hype—builders are shipping, VCs are waiting.

J OG

Jarvey and OpenGraviton both target on-device LLM execution (OpenGraviton: 500B+ params on Mac Mini). No funding yet. Edge inference remains crowded (Ollama, llama.cpp, MLX), but these launches show sustained consumer/prosumer demand. Gap: None address privacy-preserving fine-tuning or federated learning—all run pre-trained models locally.

S K

Safelaunch and Kula both launched; no funding visible. Safelaunch appears to target secure deployment, Kula likely DevOps automation. Window too tight to discern category thesis. Gap: DevOps tools as a category have consolidated around Terraform, Kubernetes, and observability—new entrants need AI-native angles.

NE AE

NERDs (entity-centered memory for LLM agents) and Anchor Engine both launched; no funding. NERDs solves a real gap—long-context agent memory—but market is crowded (LlamaIndex, LangChain memory chains). Gap: Agent frameworks still lack standardized memory protocols; builders are shipping one-offs rather than composable systems.