Connect Lever to AI Agents: Orchestrate Job Postings and Sourcing

You want to connect Lever to an AI agent so your system can autonomously draft job postings, screen candidate resumes, schedule interviews, and transition opportunities across pipeline stages. Here is exactly how to do it using Truto's /tools endpoint and SDK, bypassing the need to build and maintain a custom Applicant Tracking System (ATS) connector from scratch.

Giving a Large Language Model (LLM) read and write access to your recruiting stack is a high-stakes engineering challenge. If your team uses ChatGPT, check out our guide on connecting Lever to ChatGPT, or if you are building on Anthropic's models, read our guide on connecting Lever to Claude. For developers building custom autonomous workflows, you need a programmatic way to fetch these tools and bind them directly to your agent framework.

This guide breaks down exactly how to fetch AI-ready tools for Lever, bind them natively to an LLM using function calling using frameworks like LangChain, LangGraph, CrewAI, or the Vercel AI SDK, and execute complex talent acquisition workflows. For a deeper dive into the architectural principles of providing APIs to agents, refer to our research on Architecting AI Agents: LangGraph, LangChain, and the SaaS Integration Bottleneck.

The Engineering Reality of the Lever API

Giving an LLM access to an ATS sounds simple during a hackathon. You wrap a standard REST endpoint in an @tool decorator and let the model figure it out. In production, this approach shatters against the idiosyncrasies of enterprise SaaS APIs.

Lever's API introduces specific integration challenges that require strict defensive programming. If you hardcode these interactions into your agent's system prompt, you are forcing the LLM to memorize vendor quirks instead of reasoning about hiring workflows.

The Opportunity vs. Contact Abstraction

Unlike simpler CRMs where a "Person" applies for a "Job", Lever utilizes a deeply nested data model separating Contacts from Opportunities. A Contact is the individual human being. An Opportunity is that individual's specific journey through a pipeline for a specific role.

When an LLM attempts to advance a candidate to the "Offer" stage, it naturally tries to "update the candidate." Lever will reject this. Stages are tied exclusively to Opportunities. Your agent must know to query list_all_lever_opportunities, extract the opportunity_id, and execute a move on that specific entity. Without a unified tool schema guiding the LLM, it will hallucinate API calls against the wrong resource.

Upstream Deprecations and Epoch Timestamps

Lever's API handles time using epoch milliseconds, not standard ISO 8601 strings. Standard LLMs will frequently attempt to format dates as YYYY-MM-DD. Furthermore, Lever has deprecated the /applications endpoints in favor of querying the /opportunities endpoint with an expand=applications parameter. If you build a custom connector based on outdated documentation, your agent will immediately fail on deprecated routes.

Unforgiving Rate Limits

Recruiting APIs endure massive burst traffic. When you execute an automated resume parsing workflow, you will hit Lever's rate limits instantly.

Here is a crucial architectural reality: Truto does not retry, throttle, or apply backoff on rate limit errors. When the upstream Lever API returns an HTTP 429 (Too Many Requests), Truto passes that error directly to the caller.

However, Truto normalizes the chaotic upstream rate limit information into standardized IETF headers:

• ratelimit-limit: The total requests allowed in the current window.
• ratelimit-remaining: The number of requests left.
• ratelimit-reset: The time at which the window resets.

Your agent framework is strictly responsible for interpreting the ratelimit-reset header, sleeping the execution thread, and retrying. If you do not build backoff into your agent loop, your autonomous workflow will crash.

Why a Unified Tool Layer Matters for Agent Safety

Direct API tools - writing one Python function per raw Lever endpoint - push vendor-specific quirks into the LLM's context window. The model has to remember that API-created postings skip approval chains, that stages require specific IDs rather than text names, and that resumes expire.

Every quirk is a hallucination risk.

A unified tool layer maps internal API definitions into standardized Resources and Methods. Truto handles the authentication, pagination, and query parameter processing, exposing a clean Proxy API.

When you call the Truto /integrated-account/:id/tools endpoint, it returns a precise JSON schema for every Method defined on a Resource. Your agent sees create_a_lever_posting with a strict validation schema requiring text, categories, and location. If the LLM invents a non-existent parameter, the tool schema rejects it deterministically before the network request ever fires.

Hero Tools for Lever AI Agents

Below are the highest-leverage tools available for orchestrating Lever via Truto. Do not expose your agent to generic CRUD endpoints it doesn't need; filter your tool requests to specific, actionable operations.

list_all_lever_opportunities

This is the core reconnaissance tool for your agent. It returns pipeline opportunities, current stages, origins, and timestamps. It is the mandatory first step before attempting to update a candidate's status or pull their resume.

"Find all active opportunities in the pipeline that were added in the last 48 hours and identify which stage they are currently sitting in."

lever_opportunities_move

Moving candidates across stages is the primary write action in ATS automation. This tool requires the specific opportunity_id and the destination stage ID. It returns a note ID confirming the stage change.

"The engineering team approved candidate ID 987654321. Move their opportunity to the 'Technical Interview' stage."

create_a_lever_posting

This tool enables autonomous job creation. It requires the job description text, team, department, and location. Note that API-created postings automatically skip the human approvals chain and cannot be marked confidential.

"Draft a new job posting for a Senior DevOps Engineer based in New York. Put it under the Engineering department and leave the state as a draft so a human can review it."

list_all_lever_resumes

Before an AI agent can evaluate a candidate, it needs their professional history. This tool retrieves not only the file metadata but Lever's parsedData object, which contains extracted positions and schools, saving you from writing custom OCR logic.

"Pull the parsed resume data for the candidate associated with opportunity ID 11223344. Summarize their last three roles and extract their education history."

list_all_lever_interviews

Agents need visibility into scheduling context to prevent double-booking or to trigger feedback loops. This tool lists interviews on an opportunity, including the panel, subject, interviewers, date, and duration.

"Check if any interviews are scheduled for opportunity ID 55667788 next week. If so, return the names of the interviewers on the panel."

lever_postings_application_questions

When an agent is building a custom careers portal or syncing data to an external form builder, it must know exactly what questions Lever requires for a specific job posting. This tool returns the form fields, descriptions, and whether they are required.

"Retrieve the custom application questions for the Senior Frontend Developer posting so we can generate a matching application form on the external landing page."

To view the complete inventory of available Lever tools, including schemas for notes, feedback forms, and requisition fields, visit the Lever integration page.

Building Multi-Step Workflows

Connecting a single tool is trivial. Orchestrating a multi-step workflow where an agent discovers a state, reacts to it, handles rate limit headers, and executes a chain of API calls requires a robust framework setup.

The following example uses TypeScript, LangChain, and the truto-langchainjs-toolset. It demonstrates fetching the tools dynamically via the API and running an autonomous agent loop.

import { ChatOpenAI } from "@langchain/openai";
import { AgentExecutor, createOpenAIFunctionsAgent } from "langchain/agents";
import { ChatPromptTemplate, MessagesPlaceholder } from "@langchain/core/prompts";
import { TrutoToolManager } from "truto-langchainjs-toolset";
 
async function runLeverAgent() {
  // 1. Initialize Truto Tool Manager with your Lever Integrated Account ID
  const toolManager = new TrutoToolManager({
    apiKey: process.env.TRUTO_API_KEY,
    accountId: process.env.LEVER_INTEGRATED_ACCOUNT_ID,
  });
 
  // 2. Fetch all Lever tools dynamically from the /tools endpoint
  console.log("Fetching Lever tools from Truto...");
  const tools = await toolManager.getTools();
  
  // 3. Initialize the LLM
  const llm = new ChatOpenAI({
    modelName: "gpt-4o",
    temperature: 0,
  });
 
  // 4. Define the Agent's system instructions
  const prompt = ChatPromptTemplate.fromMessages([
    [
      "system",
      "You are an autonomous recruiting operations agent. You manage job postings, parse candidate data, and transition pipeline stages in Lever. Always verify IDs before moving opportunities."
    ],
    ["human", "{input}"],
    new MessagesPlaceholder("agent_scratchpad"),
  ]);
 
  // 5. Create and execute the agent
  const agent = await createOpenAIFunctionsAgent({
    llm,
    tools,
    prompt,
  });
 
  const executor = new AgentExecutor({
    agent,
    tools,
    maxIterations: 10,
  });
 
  try {
    const result = await executor.invoke({
      input: "Find all active opportunities. If any candidates are in the 'New Lead' stage, pull their resume data, evaluate if they have React experience, and if so, move their opportunity to the 'Screening' stage.",
    });
    console.log("Agent Execution Complete:", result.output);
  } catch (error: any) {
    // CRITICAL: Handle 429 Too Many Requests explicitly.
    // Truto passes upstream rate limits directly to you.
    if (error?.status === 429) {
      const resetTime = error.headers['ratelimit-reset'];
      console.error(`Rate limited by Lever. Resume operations after timestamp: ${resetTime}`);
      // Implement your custom sleep/backoff logic here
    } else {
      console.error("Agent execution failed:", error);
    }
  }
}
 
runLeverAgent();

The Architecture Behind the Execution

When the agent runs, it does not communicate with Lever directly. It communicates with the Truto Proxy API. This isolates the agent from authentication rot and complex query parameters.

sequenceDiagram
    participant Agent as AI Agent
    participant Truto as Truto Proxy API
    participant Lever as Lever Upstream API
    
    Agent->>Truto: Call list_all_lever_opportunities
    Truto->>Lever: GET /v1/opportunities (with Bearer Token)
    Lever-->>Truto: 200 OK (JSON payload)
    Truto-->>Agent: Returns normalized JSON
    
    Agent->>Truto: Call list_all_lever_resumes (id: 123)
    Truto->>Lever: GET /v1/opportunities/123/resumes
    Lever-->>Truto: 200 OK (Parsed Resume Data)
    Truto-->>Agent: Returns normalized JSON

Workflows in Action

To understand the true utility of AI-driven ATS orchestration, we must look at concrete, multi-step workflows. These are not simple read operations; they are autonomous business processes.

Workflow 1: The Autonomous Resume Screener

Recruiting teams waste hundreds of hours manually opening PDFs to check for basic qualifications. An AI agent can process inbound applicants in real-time.

"Check the Lever pipeline for any opportunities currently sitting in the 'Applied' stage for the Data Scientist role. For each opportunity, pull their resume. If the parsed resume data indicates less than 3 years of experience with Python and SQL, move the opportunity to the 'Rejected' stage and add a note explaining the lack of technical requirements. If they meet the requirements, move them to 'Manager Review'."

Agent Execution Path:

1. Calls list_all_lever_opportunities filtering for the specific posting ID and the 'Applied' stage.
2. Iterates through the results, calling list_all_lever_resumes for each opportunity_id.
3. The LLM evaluates the parsedData returned by Truto.
4. Based on the evaluation, it calls lever_opportunities_move to advance or reject the candidate.
5. Calls create_a_lever_note to append its reasoning directly to the candidate's profile.

Workflow 2: Programmatic Job Requisition Generation

When a hiring manager needs a new role opened, they typically fill out an internal Jira ticket or Google Form. An agent can bridge the gap between internal IT systems and Lever autonomously.

"We need a new posting for a Product Designer in London. Create the posting in Lever. Use standard design department categories. Leave the state as draft. Once created, fetch the default application questions for that posting so I can review them."

Agent Execution Path:

1. The agent synthesizes a job description based on its internal prompt knowledge of a "Product Designer."
2. Calls create_a_lever_posting passing the required text, team, department, and location parameters, explicitly setting the state to 'draft'.
3. Extracts the newly generated posting ID from the response.
4. Calls lever_postings_application_questions using the new ID.
5. Returns a natural language summary to the user confirming the draft is ready for review and listing the default questions the applicant will face.

Workflow 3: Stalled Candidate Auditing

Candidates left lingering in intermediate stages damage employer brand and ruin time-to-hire metrics. An agent can autonomously audit the pipeline and escalate stale records.

"Audit the Lever pipeline. Find any opportunities in the 'Interview' stage where the last updated timestamp is older than 14 days. Find out who the hiring manager is, and add an internal secret note to those opportunities flagging them as stalled."

Agent Execution Path:

1. Calls list_all_lever_opportunities filtering for the 'Interview' stage.
2. The agent calculates the time delta using the returned epoch timestamps.
3. For any opportunity exceeding the threshold, it calls get_single_lever_opportunity_by_id to fetch extended metadata.
4. Calls create_a_lever_note using the opportunity_id, setting the secret parameter to true, leaving a warning note for the talent team.

Moving Past Manual Integrations

Building an AI agent is fundamentally an exercise in state management and reasoning. Every sprint you spend parsing Lever's epoch timestamps, handling OAuth token refreshes, or trying to understand the difference between a contact and an opportunity is a sprint stolen from improving your core model.

By leveraging Truto's /tools endpoint, you offload the entire infrastructure burden of API integration. Your agent frameworks interact with a stable, documented, and deterministic unified layer. You retain complete control over the retry logic and execution flow, while Truto handles the chaotic reality of upstream SaaS communication.

Stop writing defensive API wrappers. Give your agents the schemas they need, and let them get to work.