June 6, 2026 · 3 min read
Sharper web search and sandboxed code execution
TaskForceAI improved web search depth and source visibility and runs code execution in isolated sandboxes, giving agents stronger research and compute tools.
Deeper web search with visible, traceable sources
Sandboxed code execution for real computation
Both tools available to agents during a run
Better answers need better inputs
Two tools do a lot of the heavy lifting in a good agent run: the ability to look things up, and the ability to compute. This week we sharpened both.
Web search now goes deeper and surfaces its sources, so research-heavy answers show where the information came from instead of asking you to take them on faith. We also improved how current news gets synthesized into a usable answer.
Run the code, do not just describe it
Code execution runs in an isolated sandbox, so an agent can actually execute code to test an idea, transform data, or check a result rather than only reasoning about it. We moved sandboxing onto Daytona for a consistent, disposable execution environment.
- Web search with greater depth and source visibility
- Sandboxed code execution on disposable environments
- Stronger synthesis of current information