ModelsJuly 31, 2026

Perplexity open sources numbat to monitor risky ai coding agents

Perplexity has released Numbat, an open-source tool designed to monitor the activities of AI coding agents on local endpoints. The utility provides detection and opt-in blocking capabilities to prevent unauthorised or dangerous actions by autonomous systems. This follows growing security concerns regarding the direct access AI agents have to development environments.

Why it matters

For enterprise teams deploying autonomous agents, Numbat offers a layer of observability and safety to mitigate risks like data exfiltration. It addresses the critical trust gap between agentic capabilities and strict enterprise security requirements.

Key points

Numbat monitors agent activity on endpoints to detect potentially malicious behaviour
It provides developers with the ability to opt in to blocking specific risky actions
The tool is open source to allow for community contributions and transparent security auditing

More AI news

Models

Anthropic releases Claude Haiku 5.5 with 75 percent cost reduction

Anthropic has launched Claude Haiku 5.5, positioning it as its fastest, cheapest, and most capable small model to date. The update cuts pricing by around 75 percent compared to earlier benchmarks for its lightweight tier. It aims to deliver rapid inference for high-volume tasks without sacrificing overall model performance.

Models

Mistral releases public preview of 1-trillion-parameter ML4 model

French artificial intelligence firm Mistral has launched a public preview of its Mistral Large 4 model. The new release features a 1-trillion-parameter multimodal architecture designed to tackle complex technical workloads. Early disclosures highlight strong performance across software engineering, coding, and cybersecurity domains.

Tooling

Google open-sources EmbeddingGemma 2 for multimodal search on edge devices

Google has open-sourced EmbeddingGemma 2, a compact multimodal embedding model that operates within less than 600MB of memory. The lightweight model enables unified search across text, code, images, video, and audio using a single query. It is built to run locally on mobile hardware, allowing complete offline search functionality.

Ready to build something
extraordinary?

15 minutes. No pitch deck. Just a conversation about what AI can do for your team.

Talk directly with our AI specialists

15 min, no strings

No sales pressure

Prototype in 7 days