Proposal Generator (RAG)

Overview

Property	Value
Workflows	3 (2 parent + 1 sub-workflow)
Triggers	HTTP (2 parent workflows) + sub-workflow
Integrations	Anthropic (Claude), Pinecone (vector DB)
Credential types	FLEXCRED (Anthropic), INSTCRED (Pinecone)
Features	RAG retrieval, file upload/download, sub-workflows, data ingestion, knowledge base
Cost	15 + 5 + 0 credits
Complexity	Complex

This use case generates business proposals by searching a knowledge base of historical proposals (via RAG), then using Claude to draft a new proposal based on the requirements document and similar past work.

Folder structure

proposal-generation/
  config.ts
  version.json
  project.json
  workflows/
    proposal-generation-package.json    # Main: generates proposal
    document-retrieval.json             # RAG: finds similar documents
    markdown-to-pdf-subworkflow.json    # Sub-workflow: converts to PDF

Workflow architecture

User uploads requirements document
  → proposal-generation-package (HTTP trigger)
    → Extract text from document
    → Search knowledge base (Pinecone) for similar proposals
    → Generate proposal with Claude (using similar docs as context)
    → Call pdf-converter sub-workflow
      → markdown-to-pdf-subworkflow
        → Convert markdown to PDF
        → Upload PDF via Codika Upload File (with execution metadata override)
        → Return documentId to parent
    → Codika Submit Result (with PDF documentId)

User can also:
  → document-retrieval (HTTP trigger)
    → Search knowledge base without generating a proposal
    → Return matching documents

config.ts highlights

Three workflow definitions

export const WORKFLOW_FILES = [
  'workflows/proposal-generation-package.json',
  'workflows/document-retrieval.json',
  'workflows/markdown-to-pdf-subworkflow.json',
];

Main workflow (HTTP trigger with file upload input)

{
  workflowTemplateId: 'proposal-generation-package',
  workflowId: 'proposal-generation-package',
  workflowName: 'Generate Proposal',
  integrationUids: ['anthropic'],
  triggers: [{
    triggerId: crypto.randomUUID(),
    type: 'http' as const,
    url: webhookUrl,
    method: 'POST' as const,
    title: 'Generate Proposal',
    inputSchema: getProposalInputSchema(),
  }],
  outputSchema: getProposalOutputSchema(),
  cost: 15,
}

File input schema

function getProposalInputSchema(): FormInputSchema {
  return [
    {
      type: 'section',
      title: 'Proposal Configuration',
      collapsible: false,
      inputSchema: [
        {
          key: 'requirements_file',
          type: 'file',
          label: 'Requirements Document',
          description: 'Upload the RFP or requirements document',
          required: true,
          maxSize: 50 * 1024 * 1024,
          allowedMimeTypes: ['application/pdf', '.docx', '.doc'],
        },
        {
          key: 'proposal_language',
          type: 'select',
          label: 'Proposal Language',
          required: true,
          defaultValue: 'french',
          options: [
            { value: 'french', label: 'French' },
            { value: 'english', label: 'English' },
            { value: 'dutch', label: 'Dutch' },
          ],
        },
      ],
    },
    {
      type: 'section',
      title: 'Additional Context',
      collapsible: true,
      inputSchema: [
        {
          key: 'notes',
          type: 'array',
          label: 'Additional Notes',
          description: 'Extra context to include in the proposal',
          itemField: { type: 'text', maxLength: 10000, rows: 6 },
          minItems: 0,
          maxItems: 10,
        },
      ],
    },
  ];
}

File output schema

function getProposalOutputSchema(): FormOutputSchema {
  return [
    {
      key: 'proposal_pdf',
      type: 'file',
      label: 'Generated Proposal',
      description: 'PDF proposal document',
    },
    {
      key: 'executive_summary',
      type: 'text',
      label: 'Executive Summary',
    },
  ];
}

Sub-workflow definition

{
  workflowTemplateId: 'markdown-to-pdf-subworkflow',
  workflowId: 'markdown-to-pdf-subworkflow',
  workflowName: 'Markdown to PDF Converter',
  integrationUids: [],
  triggers: [{
    triggerId: crypto.randomUUID(),
    type: 'subworkflow' as const,
    title: 'Convert Markdown to PDF',
    inputSchema: [
      { key: 'markdownContent', type: 'string' },
      { key: 'fieldKey', type: 'string' },
      { key: 'fileName', type: 'string' },
      { key: 'executionId', type: 'string' },
      { key: 'executionSecret', type: 'string' },
      { key: 'docTitle', type: 'string' },
      { key: 'docSubtitle', type: 'string' },
    ],
    calledBy: ['proposal-generation-package'],
  }],
  outputSchema: [],
  cost: 0,
}

Knowledge base access

knowledgeBaseAccess: {
  processDocTags: ['proposal', 'rfp'],
  processInstanceDocTags: ['proposal'],
}

This enables the workflow to access tagged documents from both the process-level and instance-level knowledge bases.

Data ingestion configuration

For embedding documents into the vector store:

export function getDataIngestionConfig(): ProcessDataIngestionConfigInput {
  return {
    workflowTemplateId: 'proposal-embedding-ingestion',
    workflowName: 'Embedding Ingestion',
    n8nWorkflowJsonBase64: loadAndEncodeWorkflow(__dirname, 'workflows/embedding-workflow.json'),
    webhooks: {
      embed: '{{PROCDATA_PROCESS_ID_ATADCORP}}/embed',
      delete: '{{PROCDATA_PROCESS_ID_ATADCORP}}/embed-delete',
    },
    purpose: 'Embed KB documents into Pinecone for RAG retrieval',
    cost: 2,
  };
}

Key patterns demonstrated

1. Multi-workflow architecture

Three workflows with clear separation of concerns — main generation, retrieval, and PDF conversion.

2. Sub-workflow with file upload

The PDF sub-workflow receives execution metadata from the parent and uses executionIdOverride / executionSecretOverride on the Codika Upload File node.

3. SUBWKFL placeholder

Parent calls sub-workflow via:

"value": "{{SUBWKFL_markdown-to-pdf-subworkflow_LFKWBUS}}"

4. File input and output

Users upload a document (requirements file), the workflow processes it, generates a PDF, uploads it via Codika Upload File, and returns the documentId as a file type output field.

5. RAG data ingestion

Separate data ingestion workflow embeds documents into Pinecone. Deployed independently via:

codika deploy process-data-ingestion ./proposal-generation

6. Knowledge base tags

Documents are filtered by tags, allowing fine-grained access control over which documents the workflow can read.

7. Cost differentiation

Main workflow costs 15 credits, retrieval costs 5, sub-workflow costs 0 (attributed to parent).

Deploy sequence

# 1. Validate
codika verify use-case ./proposal-generation

# 2. Deploy use case (all 3 workflows)
codika deploy use-case ./proposal-generation

# 3. Deploy data ingestion (separate)
codika deploy process-data-ingestion ./proposal-generation

# 4. Test retrieval
codika trigger document-retrieval --poll --payload-file - <<'EOF'
{"query": "software development proposal"}
EOF

# 5. Test full generation (with file upload via platform UI)
# File uploads require the platform UI — CLI trigger only supports JSON payloads

Getting Started

Why Codika

Core Concepts

Guides

Builder System

Examples

Overview

Folder structure

Workflow architecture

config.ts highlights

Three workflow definitions

Main workflow (HTTP trigger with file upload input)

File input schema

File output schema

Sub-workflow definition

Knowledge base access

Data ingestion configuration

Key patterns demonstrated

1. Multi-workflow architecture

2. Sub-workflow with file upload

3. SUBWKFL placeholder

4. File input and output

5. RAG data ingestion

6. Knowledge base tags

7. Cost differentiation

Deploy sequence

​Overview

​Folder structure

​Workflow architecture

​config.ts highlights

​Three workflow definitions

​Main workflow (HTTP trigger with file upload input)

​File input schema

​File output schema

​Sub-workflow definition

​Knowledge base access

​Data ingestion configuration

​Key patterns demonstrated

​1. Multi-workflow architecture

​2. Sub-workflow with file upload

​3. SUBWKFL placeholder

​4. File input and output

​5. RAG data ingestion

​6. Knowledge base tags

​7. Cost differentiation

​Deploy sequence

Overview

Folder structure

Workflow architecture

config.ts highlights

Three workflow definitions

Main workflow (HTTP trigger with file upload input)

File input schema

File output schema

Sub-workflow definition

Knowledge base access

Data ingestion configuration

Key patterns demonstrated

1. Multi-workflow architecture

2. Sub-workflow with file upload

3. SUBWKFL placeholder

4. File input and output

5. RAG data ingestion

6. Knowledge base tags

7. Cost differentiation

Deploy sequence