Get started

To apply for IBM Agent Connect, fill out the sign up form in the IBM Agent Connect page.

After you fill out the form, you will receive a welcome letter from IBM by email (from the IBMAgentConnect sender), and IBM representatives will set up a initial meeting to discuss the company’s use case, domain, and fit within the program.

When you are accepted into the program, you must sign a Catalog Listing Agreement (CLA) that outlines the terms and conditions for listing the company’s products or services on IBM’s platforms, including revenue share, logo rights, and pass-through terms.

The onboarding process includes:

• Onboard training
• Setting up a vendor ID
• Generating part numbers
• Establishing pricing plan and metrics
• Support linkage
• Service Brokerage for provisioning and metering

To ensure a secure and reliable interaction with watsonx Orchestrate, the following requirements must be met:

• Security and authentication
  The service endpoint must be secured with HTTPS.
  • Authentication is supported via:
    • API Key
    • Bearer Token
    • Additional authentication methods will be supported in future releases.
• API contract: External agents must adhere to the watsonx Orchestrate API contract, which is closely aligned with the OpenAI assistant specification, and follows the Agent Connect Framework specification.
• Performance requirements
  • Preferred Latency:
    • Time To First Byte (TTFB): under 6 seconds for 95% of requests.
    • Total response time: under 24 seconds.
• Request and response format
  • The service must accept structured request bodies.
  • Responses must be emitted as Server-Sent Events (SSE), compliant with the watsonx Orchestrate specification.
• Memory and state requirements
  • Statelessness: External agents must be stateless, with all necessary context for a given interaction included in each request.
  • No reliance on prior state or stored memory between requests is allowed.
• Input Handling and Reasoning
  • Agents must handle ambiguous or conflicting inputs using:
    • Context
    • Clarification prompts
  • Fabrication of information is not permitted.
  • Agents should avoid:
    • Excessive reasoning loops
    • Recursive calls
  • Control must be returned to the user within a clearly defined step or recursion limit.

1. You must test your agents using the watsonx Orchestrate Agent Development Kit testing toolkit (Coming soon).

2. To run the tests, you must use the following command, providing an input file following the test specifications (Coming soon):

   Copy

   Ask AI

   orchestrate evaluations external_validate --config external_agent.yaml --input input_file.csv
   

3. Once the tests are completed, the tool generates the test results.

4. Compress the test results into a zip file, and fill out the following YAML file with your self-certification results:

   Copy

   Ask AI

   self_certification:
       agent_name: "my_agent_name"
       partner_name: "My Company Ltd."
       test_run:
           passed: true
           test_suite_version: "1.2.1"  # ADK test suite version
           total_tests: 18  # tests run by the ADK test suite
           passed_tests: 18  # tests completed
           test_report_filename: "test_results.zip" # name of the zip file with the test results
       signed_by: "mycompany@example.com"
       certified_on: "2025-06-09T08:45:00Z"
       notes: "All tests passed successfully. Report attached."
       contact_email: "techlead@acme.io"
   

   See all 13 lines

5. Create a catalog_metadata.yaml file to submit for approval:

   catalog_metadata.yaml

   Copy

   Ask AI

   name: "box_file_comments_agent"
   version : "1.8"
   display_name: "File comments management"
   category: "agent"
   kind: "external"
   provider: "partner_name"
   description: "An agent to manage file comments in Box, with capabilities to add, update, retrieve, and delete comments."
   tags:
     - "Productivity"
   api_url: "https://api.mycompany.example.com/v1"
   auth_config: {token: "mytoken1234567890"}
   auth_scheme: "Bearer-token" # Or API Key
   chat_params: # any implementation specific parameters
       model: "granite-3.3-3b"
       store: true
   config: {}
   

   See all 16 lines

6. Create a self_attestation.yaml file and fill out the requirements appropriately:

   self_attestation.yaml

   Copy

   Ask AI

   self_attestation:
       agent_name: "your-agent-name"
       partner_name: "your-company-name"
       attested_by: "your.name@company.com"
       attested_on: "2025-06-10T00:00:00Z"
   agent_requirements:
       meets_core_functionality: true      # Agent meets the behavior and quality requirements listed
   listing_created: true                   # A listing has been created for this agent
   part_number_linked: true                # The listing is linked to a valid IBM Cloud Catalog part number
   orchestrate_association: true           # The part number is associated with watsonx Orchestrate
   trust_and_safety:
       security_best_practices: true       # Follows industry best practices for security and privacy
       ibm_security_by_design: true        # Complies with IBM Security and Privacy by Design principles: https://www.ibm.com/trust/security-and-privacy-design
       trustworthy_ai_principles: true     # Aligns with IBM’s Trustworthy AI principles: https://www.ibm.com/think/topics/trustworthy-ai
       limitations_disclosed: true         # Limitations and intended use are clearly communicated
   notes: "All requirements have been reviewed and satisfied to the best of our knowledge."
   

Use this checklist to ensure all requirements are met before submitting your agent for review.

1. Agent Requirements

   • Agent meets all functional and technical requirements.
   • A unique listing is created for each agent.
   • Each listing is linked to a corresponding part number in the IBM Cloud Catalog.
   • The part number is correctly associated with watsonx Orchestrate.

2. Security and Privacy Compliance

   • Follows industry best practices for security and privacy.
   • Complies with IBM Principles for Security and Privacy by Design.
   • Aligns with IBM’s Trustworthy AI Principles, including:
   • Fairness
   • Transparency
   • Explainability

3. Documentation and Communication

   • Clearly communicates the agent’s limitations and intended use.

4. Submission Package
   Ensure the following artifacts are included in the submission email to your IBM sales representative:

   • self_attestation.yaml file
   • catalog_metadata.yaml
   • test_results.zip file (if applicable)
   • Enablement materials (e.g., user guides, onboarding content)

Ensure that your submission includes the following key components:

1. Agent and product overview

   • Provide a clear, concise description of your agent and product.
   • Highlight core features, functionalities, and use cases.

2. Value proposition

   • Explain the unique benefits your agent/product delivers.
   • Focus on how it solves customer pain points or improves outcomes.

3. Competitive differentiation/positioning

   • Outline what sets your solution apart from competitors.
   • Include any unique technology, pricing advantages, or market positioning.

Make sure all materials are professional, easy to understand, and tailored to IBM’s audience.

Users acquire agents through the watsonx Orchestrate Agent Catalog, and your company receives payments according to the terms defined in the agreement.

Your company is responsible for maintaining your Agent up to date and submit new versions to IBM for publication in the Catalog.