AI agent builder

Build AI agents that decide.
Keep the final say.

Fetch Hive is a no-code AI agent builder. Give an agent a model, instructions, tools, and knowledge. It picks the right tool for each request, calls your workflows and connected apps, remembers the thread, and pauses for your approval before anything with side effects. Test it in chat, then run it from your code or the AI assistant you already work in.

Free Developer plan ยท No card required ยท Agent templates you can clone in one click

Fetch Hive Workflow Editor

01

ANATOMY

Model, instructions, tools, knowledge. One AI agent you can call.

An AI agent is not a fixed path. It reads the request, decides which tool to use next, reads the result, and keeps going until it has an answer. You set the boundaries: the model, the instruction prompt, the tools it may call, and what needs your approval.

Model and instructions

01

Pick an OpenAI, Anthropic, Gemini, xAI, Perplexity, or Kimi model. Set temperature or reasoning effort, max tokens, and an instruction prompt that says what the agent is for and how it should behave.

Tools

02

Add web search, code execution, image generation, or any app connected under Integrations. The agent reads each toolโ€™s description and decides when to call it.

Workflows and sub-agents

03

Hand a deployed workflow to the agent as a tool, or let one agent call another. Each call runs with its own logs and trace, nested under the parent run.

Knowledge and memory

04

Attach a knowledge base so answers draw on your own documents. Pass a thread_id and Fetch Hive keeps the conversation history for you between calls.

02

BUILD

Build AI agents without code. Test them in chat before anyone else does.

The agent editor shows the root agent and every tool as nodes on one canvas. Change a setting, add a tool, and send a test message from the same screen. Test chats are not saved, so you can iterate freely.

Agent editor

Agent editor with the root agent and tool nodes

The agent and its tools, as nodes.

Select the root node to open Agent Settings, or any tool node to configure it. Add a tool from the header and it appears on the canvas with its own settings sheet.

Tool picker

Tool picker with MCP tools and workflow tools

Built-in tools, apps, workflows, and other agents.

Search MCP tools, pick a deployed workflow from Workflow Tools, or add a sub-agent from the same workspace. Over 6,400 app actions are one connection away.

Tool picker

Tool picker with MCP tools and workflow tools

Built-in tools, apps, workflows, and other agents.

Search MCP tools, pick a deployed workflow from Workflow Tools, or add a sub-agent from the same workspace. Over 6,400 app actions are one connection away.

Approvals

Tool approval request waiting for a decision

Nothing with side effects runs without a yes.

Mark the tools that need a human. A run pauses as awaiting approval, you allow once or deny from the dashboard or the API, and it continues on the same thread.

Approvals

Tool approval request waiting for a decision

Nothing with side effects runs without a yes.

Mark the tools that need a human. A run pauses as awaiting approval, you allow once or deny from the dashboard or the API, and it continues on the same thread.

Test in chat

Chat panel testing an agent in the editor

Talk to it before you ship it.

Open Chat in the editor header and send a message. You see the reply, the tools it called, and the tokens it used. Nothing is saved, so you can push on it until it behaves.

03

DEPLOY

Call your AI agent from your app. Pick up the thread next time.

Every saved agent has one invoke endpoint. Your code, a webhook from another service, or your assistant over MCP all talk to the same agent, and a thread_id turns a single message into a conversation.

Threads

Pass any string as thread_id, such as a user id or a ticket number. Fetch Hive creates the thread on the first call and resumes it on every call after, with the history included.

Webhooks

Let an external service start an agent run with the agentโ€™s own webhook secret instead of your API key. The reply is delivered to your callback URL and signed.

API and SDKs

Invoke an agent with your API key from cURL or the Python, Node.js, Ruby, and PHP SDKs. Stream the response as it is generated, or wait for the final JSON.

Approvals

When a tool needs approval, the response comes back as awaiting approval with a review link. Allow once or deny from the dashboard or the API, and the run continues.

agent invoke ยท support-triage

# send a message to the agent from any system that can POST

$ curl https://api.fetchhive.com/v1/agent/invoke \

-H โ€˜Authorization: Bearer <api key>โ€˜ \

-d โ€˜{ โ€œagentโ€: โ€œ<agent id>โ€, โ€œthread_idโ€: โ€œticket-42โ€,

โ€œmessageโ€: โ€œSummarize this ticket and draft a replyโ€ }โ€™

โœ“ req_01HZโ€ฆ streaming ยท thread ticket-42 resumed with 3 earlier turns

Status

awaiting approval

Thread

ticket-42 ยท history loaded by Fetch Hive

Tools called

search_tickets ยท knowledge_search ยท draft_reply (needs approval)

Next

allow once or deny, then the run continues

THREADS AND APPROVALS

A thread is the memory of a conversation, under any id you choose. An approval is a pause the agent cannot skip. Together they let an agent work on your behalf for days without ever acting without you.

Reuse a thread_id and the agent picks up where it left off

Or send the earlier turns yourself and nothing is stored

Approvals expire, carry a review link, and are logged with who decided

Always allow stays in the dashboard, never in an API call

04

INTEGRATE

Give one AI agent 6,400+ tools and let it choose.

Connect your apps once under Integrations and add them to any agent. 73 live providers, more than 6,400 agent tools, and every call the agent decides to make is recorded in its run trace.

6,400+

AGENT TOOLS

Actions an agent can call today

73

LIVE PROVIDERS

Apps connected once, reused by every agent

6

MODEL PROVIDERS

OpenAI, Anthropic, Gemini, xAI, Perplexity, and Kimi

Customer email

Message in

Support agent

Decides and acts

Zendesk

Ticket in Zendesk

Tool call

Slack

Reply drafted

Awaiting approval

05

CONNECT

Talk to your AI agents from the assistant you already use.

Fetch Hive ships an MCP server. Connect Claude, ChatGPT, Cursor, or Codex once, and your assistant can find, build, configure, and run agents in plain language. Every run is logged like any other.

What your assistant can do once it is connected

Find and read any agent in your workspace, tools and knowledge included

Send a message with the cost quoted first, then approve it in the chat

Build a new agent from a sentence, or clone and tweak a template

Change the model, instructions, tools, or knowledge, with undo

Approve or deny a paused tool call from the chat

Open the run trace and the reply without leaving the conversation

WORKS WITH

Google Gemini
Claude Code
Google Gemini
Claude Desktop
ChatGPT
Codex
Cursor
Visual Studio Code
VS Code
Google Gemini
Gemini CLI
Google Gemini
JetBrains

Sign in once with OAuth. Nothing runs without your yes.

06

OBSERVE

Every agent run leaves a trace you can open.

Agent Logs keep each run with its model, tokens, tasks, tool calls, and the full trace, with sub-agent and workflow calls nested under the parent. Filter by provider, model, API key, or your own metadata.

01

The task

Produce a company research brief with sources. The input is a company URL; the expected output is a one-page document your team can read.

02

The execution

Every model call, tool call, duration, and recorded cost. Open a meaningful step to see its input and output.

03

The output

A readable excerpt from the resulting brief, with sources where supported, next to the run summary.

Every run records:

Every message in and out, with the thread it belongs to

Each tool call, its arguments, result, and duration

Tokens and cost, broken down by parent agent, sub-agents, and tools

Approval decisions, who made them, and when

Value is a human judgement. The run gives you every decision the agent made and the recorded cost of making it, and the same trace is what your assistant reads over MCP.

07

START FASTER

Clone an AI agent template, or create an AI agent from a sentence.

Most agents start as a copy of one that already works. Or describe what you need and Fetch Hive drafts the agent, its instructions, and its tools for you to review.

Templates ยท Free to clone

I WANT TO

Start from something that already works.

Browse agents built by Fetch Hive and the community. Clone one into your workspace, swap the model or the tools, and it is yours.

Filter the marketplace to agents only

Keeps the instructions and built-in tools; reconnect apps in your workspace

Every template shows how many people use it

No account needed to browse

Describe it

I WANT TO

Describe the agent and let Fetch Hive draft it.

Tell Fetch Hive what the agent should do, in the dashboard or from your AI assistant. It asks what it needs to know, drafts the instructions, picks the tools, and saves a draft you can test right away.

Answer a few questions and review the draft before anything runs

Test it in chat, then invoke it from your code

Change anything later in the editor

Free Developer plan

MOST USED AGENT TEMPLATES

PART OF THE FETCH HIVE PLATFORM

An AI agent platform that shares your connected apps, knowledge bases, and deployed workflows.

Connect Slack, Zendesk, Google, HubSpot, and the rest once under Integrations, and every agent can use them. The same workspace runs your workflows, so an agent can hand a task to one as a tool.

FAQ

Questions people ask before their first agent.

Everything else is in the agent docs.

An AI agent builder is a tool for creating AI agents without writing the orchestration code yourself. In Fetch Hive you pick a model, write the instructions, and add tools, knowledge, and approvals in a visual editor. No code is needed to build or test an agent, and the same agent can be called from the API when you are ready.