Koder now builds and deploys your agents. Describe the agent; Koder writes, tests and deploys it, and you review every change.

View on GitHubDownload for desktop

aixplain Agents SDK

The fastest way to turn an AI agent into an API. Deploy agents with one save, embed them in your apps, and pay only when someone calls them. Use any of 1,000+ models, tools and integrations with one API key, cap cost with budgets, check every step with inspectors, and trace every run in code or in the aixplain app dashboard.

Anatomy of an agent

Agent

Runagent.run("...")Prompt, files, session

Agentaix.Agent(...).save()

LLM

llm=

Tools

• + Slack, Gmail, MCP, Python

tools=

Subagents

• Researcher
  • Web search
• Writer

agents=

Inspectors

Check input, steps and answer

inspectors=

Budget

Cap cost, time and steps

agent.budget

Resultresult.data.outputresult.data.stepsresult.used_credits

Python

from aixplain import Aixplain

aix = Aixplain(api_key="YOUR_API_KEY")

llm = aix.Model.get("openai/gpt-5.4/openai")
web_search = aix.Tool.get("tavily/tavily-web-search/tavily")

guard = aix.Inspector(  # optional: severity, description
    name="no-competitors",
    targets=["output"],
    action="abort",
    metric={"assetId": llm.id, "prompt": "Fail if it names Globex."},
)

researcher = aix.Agent(name="Researcher", tools=[web_search])

agent = aix.Agent(  # optional: description, instructions, tools
    name="Research agent",
    llm=llm,
    agents=[researcher, aix.Agent(name="Writer")],
    inspectors=[guard],
)
agent.budget.max_cost = 0.5
agent.save(save_subcomponents=True)

result = agent.run("What are the top AI papers this week?")
print(result.data.output)

API

curl -X POST \
  "https://platform-api.aixplain.com/v2/agents/YOUR_AGENT_ID/run" \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "What are the top AI papers this week?"}'

Use agent.id from the Python tab. The response holds a URL to poll for the answer; see the REST API.

Ways to build

• Python SDKDefine, deploy and run agents in code.Quickstart
• REST API and MCPCall any agent, model or tool from any language or MCP client.REST API
• Coding agentDescribe the agent. Koder or your coding agent builds and deploys it.Build with coding agents

No code? Use the aixplain desktop app.

Recommended path

1. Quickstart: run your first agent in five minutes.
2. API keys: keep your key in AIXPLAIN_API_KEY, out of git.
3. Agents: instructions, LLM, JSON output and subagents.
4. Tools and Knowledge base: web search, apps and your documents.
5. Budgets and Inspectors: cap cost and check every step.
6. Deployment and Tracing and monitoring: call it by ID and see every run.

What you can build

Compare the top three vector databases and estimate their monthly cost.

• Web search
• Python sandbox

LLM

Answer

A comparison that cites its sources

Every tool call is in result.data.steps.

Start with Web search

What is our parental leave policy?

• Knowledge base

LLM

Answer

An answer grounded in the matching passages

Every tool call is in result.data.steps.

Start with Knowledge base

Which customers churned last quarter, and why?

• SQL database
• Knowledge base

LLM

Answer

An answer from your tables and your support notes

Every tool call is in result.data.steps.

Start with Databases

Summarize today's support emails and post it to Slack.

• Gmail
• Slack

LLM

Answer

The summary, posted in your channel

Every tool call is in result.data.steps.

Start with Integrations

Research the EV market and write a one-page brief.

• Researcher agent
• Writer agent

LLM

Answer

One brief, built by two subagents

Every tool call is in result.data.steps.

Start with Subagents

Draft a reply to this customer complaint.

• Knowledge base
• Gmail

LLM

Answer

A reply that passed your inspector, or a stop with the reasonInspector checks the reply

Every tool call is in result.data.steps.

Start with Inspectors