Red Hat AAP 2.7 introduces BYOK ("Bring Your Own Knowledge") as a Tech Preview feature, announced at Red Hat Tech Day Netherlands 2026 in Bunnik. BYOK lets you inject internal documents — runbooks, change-management procedures, network naming conventions, compliance policies — into the Intelligent Assistant's RAG pipeline, so chatbot answers reflect your organization's own standards instead of generic Ansible documentation. Below is an example playbook that uploads a runbook document into the BYOK knowledge source.

---
- name: Add a company runbook to the Intelligent Assistant BYOK knowledge base
  hosts: localhost
  gather_facts: false
  vars:
    aap_controller_host: "https://aap.example.internal"
    aap_byok_knowledge_source: "network-ops-runbooks"
    runbook_file: "./runbooks/incident-response-vlan-outage.md"

  tasks:
    - name: Ensure the BYOK knowledge source exists (Tech Preview API)
      ansible.builtin.uri:
        url: "{{ aap_controller_host }}/api/gateway/v1/byok/knowledge_sources/"
        method: POST
        headers:
          Authorization: "Bearer {{ aap_token }}"
        body_format: json
        body:
          name: "{{ aap_byok_knowledge_source }}"
          description: "Runbooks and change-management procedures for network ops"
        status_code: [201, 400]
      register: knowledge_source
      no_log: true

    - name: Upload the runbook document for RAG ingestion
      ansible.builtin.uri:
        url: "{{ aap_controller_host }}/api/gateway/v1/byok/knowledge_sources/{{ aap_byok_knowledge_source }}/documents/"
        method: POST
        headers:
          Authorization: "Bearer {{ aap_token }}"
        body_format: form-multipart
        body:
          file:
            content: "{{ lookup('file', runbook_file) }}"
            filename: "incident-response-vlan-outage.md"
            mime_type: "text/markdown"
          tags: "runbook,network,incident-response"
        status_code: 202
      register: upload_result

    - name: Show ingestion job status
      ansible.builtin.debug:
        msg: "Runbook queued for RAG ingestion, job id: {{ upload_result.json.job_id }}"

What it does and why

The play targets two Tech Preview BYOK endpoints exposed by AAP 2.7's gateway API. The first task creates (or confirms) a named knowledge source — a logical bucket for related documents, such as network-ops-runbooks or pci-dss-controls. The second task uploads a Markdown runbook (incident-response-vlan-outage.md) into that source as a multipart file; AAP queues it for chunking, embedding, and indexing into the Intelligent Assistant's vector store. Once ingestion completes, the chatbot embedded in the AAP UI can cite that runbook when a user asks about the specific incident procedure, instead of falling back to generic upstream Ansible guidance.

BYOK targets the Intelligent Assistant only — it does not extend to the Coding Assistant (the VS Code extension), which has a separate BYOM provider matrix.

Notes / Gotchas

  • BYOK is Tech Preview in AAP 2.7 as of Red Hat Tech Day Netherlands 2026 — the API paths above are illustrative of the shape of the feature; confirm exact endpoint names against your AAP 2.7 gateway API reference before running in production.
  • The Intelligent Assistant's BYOM matrix matters here: BYOK content is retrieved and passed to whichever chat model backs your Intelligent Assistant deployment (Red Hat AI, OpenAI, or Azure OpenAI on AAP 2.6+). IBM watsonx is not supported for the Intelligent Assistant, so BYOK content won't reach a watsonx-backed chatbot. Google Gemini/Vertex support for the Intelligent Assistant was listed as Coming Soon.
  • Keep runbooks in plain Markdown or plain text where possible — cleaner source formatting produces better chunking and retrieval quality in the RAG pipeline.
  • Treat compliance documents (PCI-DSS, HIPAA, SOC2 policies) as sensitive: apply the same access controls to the BYOK knowledge source as you would to the source documents themselves, since retrieved chunks can surface verbatim in chatbot answers.
  • Because this is Tech Preview, don't build hard production dependencies around it yet — expect API and behavior changes before general availability.

BYOK turns the Intelligent Assistant from a generic Ansible help bot into one that answers with your organization's own runbooks and policies, but until it graduates from Tech Preview, pin your automation to a specific AAP 2.7 point release and re-verify the API shape on each upgrade.

References