Built-in agent tools — web search, browse, images, code & artifacts
v1.24.0 gives every agent a general toolbelt out of the box: live web search, page fetch + crawl, image generation, a safe calculator, and artifact creation. Providers are pluggable; defaults work with minimal setup.
Until now a Vocion agent could reason brilliantly over your data — everything you'd ingested through Sources — but it couldn't reach past it. Ask it about this morning's news, hand it a URL, or ask for a chart and it had nowhere to go. v1.24.0 closes that gap: the general toolbelt that was on the Tools roadmap is now built in.
What's new
Every agent gets these, alongside the existing search_knowledge / run_operation toolset:
web_search— live web search. Ranked titles, URLs, and snippets for current events and external research.fetch_url— read a single page and get back clean text/markdown.crawl_site— same-origin, breadth-first crawl with capped depth and page count.generate_image— create an image/graphic from a prompt. Saved as an artifact, returned as a URL.run_code— evaluate a math expression precisely (no more arithmetic in the model's head).create_artifact— produce a downloadable CSV, SVG chart, or doc.
The agent uses search_knowledge for what you've connected, and the new tools for the live web and for producing things.
Pluggable providers, sensible defaults
Each capability resolves a provider from an env var, and degrades gracefully — if a key is missing, the tool returns a clear "not configured" message instead of failing the turn.
| Capability | Default | Optional |
|---|---|---|
| Web search | Tavily (TAVILY_API_KEY) | Brave |
| Browse | Builtin (no key) | Firecrawl, for JS-heavy pages |
| Image | OpenAI gpt-image-1 (reuses OPENAI_API_KEY) | — |
| Code | Builtin safe calculator (no key) | E2B sandbox |
| Artifacts | Builtin (no key) | — |
Only web search needs a new key for full functionality. The builtin browse provider reuses the same extractor as the web Source connector, and the builtin code provider is a safe calculator — it evaluates math against an allowlist and never runs arbitrary code. Full sandboxed execution is the opt-in E2B provider.
Where to find it
- In the app: Dashboard → Capabilities → Tools shows every tool, its active provider, and whether its key is set.
- Over MCP: all six are exposed on the MCP server, callable by any MCP client.
- Docs: Tools · Using built-in tools.
Get it
Pull v1.24.0, copy the new keys from .env.example, and your agents can search, browse, draw, calculate, and produce files — no authoring required.