Chat Context

Chat context configures what information to inject into post-pipeline interactive chat sessions. When a pipeline completes, Wave can start a chat session pre-loaded with pipeline results so users can explore outcomes conversationally.

Basic Configuration

kind: WavePipeline
metadata:
  name: audit-security

chat_context:
  artifact_summaries:
    - findings
    - recommendations
  suggested_questions:
    - "What are the critical vulnerabilities?"
    - "Which files are most affected?"
    - "What should we fix first?"
  focus_areas:
    - security
    - authentication
    - data-validation

steps:
  - id: scan
    persona: auditor
    exec:
      type: prompt
      source: "Scan for security vulnerabilities"
    output_artifacts:
      - name: findings
        path: .wave/output/findings.json
        type: json
      - name: recommendations
        path: .wave/output/recommendations.md
        type: markdown

Fields

Field	Default	Description
artifact_summaries	[]	Artifact names to summarize and inject into the chat context
suggested_questions	[]	Opening questions displayed to the user when the chat session starts
focus_areas	[]	Topic areas to highlight, helping the chat session stay relevant
max_context_tokens	8000	Token budget for injected context

Artifact Summaries

List the artifact names (from any step's output_artifacts) to include in the chat context. Wave summarizes these artifacts and injects them as background context for the chat session.

chat_context:
  artifact_summaries:
    - analysis        # from step: analyze
    - test-results    # from step: test
    - implementation  # from step: implement

Only reference artifacts that provide useful background. Large artifacts are truncated to fit within max_context_tokens.

Suggested Questions

Provide starting questions relevant to the pipeline's output. These appear as clickable suggestions when the chat session opens.

chat_context:
  suggested_questions:
    - "Summarize the key findings"
    - "What patterns emerged from the analysis?"
    - "What are the recommended next steps?"

Focus Areas

Focus areas guide the chat session toward relevant topics, reducing off-topic responses.

chat_context:
  focus_areas:
    - performance
    - api-design
    - error-handling

Token Budget

Control how much context is injected. Larger budgets provide more detail but consume more of the model's context window.

chat_context:
  max_context_tokens: 16000

Default is 8000 tokens. Set higher for complex pipelines with many artifacts, lower for simple pipelines where you want the chat session to be more responsive.

When to Use Chat Context

• Exploratory analysis: After audit or research pipelines, let users dig into findings
• Implementation review: After implementation pipelines, chat about the changes made
• Decision support: After planning pipelines, discuss recommendations interactively

See Also

• Pipeline Schema: Chat Context - Field reference
• Outcomes - Structured deliverable extraction (complementary feature)