the agent runtime
ship agents that run in production.
not a framework. a runtime. durable execution, 250+ tools, human-in-the-loop. out of the box.
what you stop building yourself
you write
we handle
durable execution
every tool call is an event. agent crashes? resumes where it left off. no lost work.
human-in-the-loop
approval gates before dangerous actions. mobile approval. always-allow per tool.
channels
agents live on slack, telegram, discord, web. not just chat, real integrations.
flows as tools
agents call composed workflows like any other tool. complex orchestration, simple interface.
triggers & crons
scheduled execution. webhooks. agents work while you sleep.
250+ tools
every model on inference.sh as a tool. plus MCP servers. plus skills. one agent, everything.
the engine under the hood
agents compose everything on the platform. tools for capabilities. skills for knowledge. MCP for connections. the runtime ties it all together.
npm i -g @inference/beltfrequently asked questions
is this a framework or a runtime?
a runtime. you don't import a library. you deploy agent logic and we execute it with durable state, retries, and observability built in.
what models can agents use?
all 250+ tools on inference.sh, plus any MCP server, plus custom skills. one agent can call FLUX for images, Claude for reasoning, and a Slack MCP server, all in the same session.
can I use agents without the rest of the platform?
yes. the agent runtime is standalone. skills, MCP, teams, and UI are optional layers on top.
we use cookies
we use cookies to ensure you get the best experience on our website. for more information on how we use cookies, please see our cookie policy.
by clicking "accept", you agree to our use of cookies.
learn more.