Agents
Ollama and Fly.io integrate for secure agent sandboxes
Ollama Bloghigh confidence
Why it matters
- →Isolates AI agent code execution in persistent VMs to prevent infrastructure compromise.
- →Eliminates API key leakage risks by using a gateway that injects credentials internally.
- →Enables safe autonomous coding workflows with tools like Claude Code and Gemini CLI.
Ollama and Fly.io have launched an integration allowing AI agents to run securely within Fly.io's Sprite virtual machines. The setup uses a gateway that injects Ollama API keys internally, ensuring agents never see the credentials directly. Supported coding agents include Claude Code, Codex, and Gemini CLI, which can now access models like kimi-k3 and glm-5.3 from within isolated environments. This architecture enables safe execution of untrusted code, such as fuzzing tests, by separating inference capabilities from command execution.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1Google Enhances Gemini API with Managed Agents
Google AI Blog · July 28, 2026 · Background
- 2Docker Sandboxes Enhance Security for AI Agents
Sam Witteveen · August 19, 2026 · Background
- 3Claude Desktop Integrates with Ollama Gateway
Ollama Blog · August 25, 2026 · Related
- 4OpenAI Launches Agents API for Cloud Agents
OpenAI · September 10, 2026 · Related
- 5Ollama adds Jev-style decision models
Ollama Blog · September 29, 2026 · Related
- 6OpenAPPA tackles agent prompt injection
Fireship · September 30, 2026 · Background
- 7GitHub Copilot CLI v1.0.94 adds local Ollama model discovery
GitHub Changelog · October 7, 2026 · Background
- 8Copilot adds sandboxing and Claude Haiku 5.5
GitHub Changelog · October 9, 2026 · Background
What happened next
- 9Meta Muse Spark Open launches on Together AI
Together AI Blog · October 15, 2026 · Background
- 10Claude Code v2.1.307 patches stability and API versioning
Claude Code Releases · October 22, 2026 · Background
- 11Claude Code v2.1.305 fixes critical sandbox and retry regressions
Claude Code Releases · October 22, 2026 · Background
- 12OpenAI trains frontier models offline after agent hacks
The Verge AI · October 23, 2026 · Background
- 13llama.cpp b11940 adds deferred tool support
llama.cpp Releases · October 28, 2026 · Background
More in Agents
© The AI Daily BriefAgentscoding
Spawn ships 77k code changes monthly with AI agents
Ten-person team at Spawn utilizes an always-on fleet of AI agents to achieve massive monthly code output.
The AI Daily Brief
© WIRED AIAgentsagents
AI Agent Performs Live Mouse Brain Injection
Exin Therapeutics has demonstrated an AI agent physically controlling a robotic arm to inject drugs into a live mouse's brain. The experiment reveals a critical divergence in model behavior: Anthropic’s Claude Opus refused the procedure due to safety concerns, while Sonnet proceeded. This suggests that as agents gain direct access to physical lab equipment, their alignment and refusal mechanisms become operational realities rather than theoretical safeguards. It marks a shift from AI merely analyzing data to actively manipulating biological systems under human supervision. The team initially tested Claude Opus 5.5, but it halted the injection, citing potential harm. Switching to Claude Sonnet 5 allowed the procedure to complete, though with delays caused by coordinate system glitches. A secondary Opus agent had to fix code errors without knowing a live surgery was underway. Human oversight remains essential as researchers test more complex tasks like craniotomies.
WIRED AI
© Hugging Face BlogAgentsagents
NVIDIA Nemotron powers interactive Halloween agent
An NVIDIA engineer built a fully local voice agent for a Halloween attraction using the new Nemotron Voice Agent framework. The system chains Nemotron ASR, the Bolt E4B small language model, and Magpie TTS to create a grumpy pirate persona that responds to trick-or-treaters in real-time. Running on dual RTX PRO Blackwell GPUs, it achieves sub-2-second latency for uncached responses, proving these open models can handle complex, multi-turn interactive tasks without cloud dependency. This demonstrates the practical viability of local AI agents for consumer-facing applications.
Hugging Face Blog