RCS AI

Stable Diffusion v3 Image

PowerPoint Generator

Mermaid Diagram

Render Chart

GPT Image Editor

DALL-E 3

generate_redline

Generate a redline-marked PDF with numbered compliance correction boxes drawn on the source ad. Call this AFTER completing the compliance analysis and producing the itemized corrections list. Requires a file_id from the pre-upload step (user uploads PDF at /upload page before starting review). Each correction must include anchor_text (exact text snippet from the ad to locate each issue). Returns a download link for the marked-up PDF.

ROAD-AI PDF Redline

generate_excel

Generate a .xlsx Excel workbook from a sheet specification. Use this when the user asks for a spreadsheet, workbook, table-of-data file, summary export, side-by-side comparison, or anything that benefits from rows-and-columns format with optional formulas. Each sheet has optional headers (rendered bold + frozen) and a list of data rows. Strings beginning with `=` are written as formulas (e.g. "=SUM(B2:B10)"). Null values produce empty cells.

Note: This is general-purpose Excel generation. For the standardized Dochterman client workbook write workflow (Policy Checking / Building Data / Prospect Research / MVR agents), a dedicated workbook endpoint will be used — not this function.

Payload schema (pass as JSON string in `payload_json`):
```
{
  "filename_hint": "Optional preferred filename stem",
  "client": "Optional client name for indexing",
  "agent": "Optional agent identifier",
  "retention_days": 14,                                  // min 14, max 90
  "sheets": [
    {
      "name": "Sheet name (max 31 chars, auto-sanitized)",
      "headers": ["Col A", "Col B", "Col C"],            // optional
      "rows": [[1, 2, 3], [4, 5, "=SUM(A2:B2)"]],         // mixed types OK
      "column_widths": [null, 15.0, 30.0]                // optional, null = auto
    }
  ]
}
```
Returns: { id, filename, download_url, expires_at, size_bytes }. The download_url is signed and surfaced in chat as a clickable link.

RCS Excel Generator

generate_word

Generate a .docx Word document from a structured block list. Use this when the user asks for a Word document, proposal, narrative, memo, summary, or letter. The agent assembles the document programmatically (no template), and the server renders it and returns a download URL.

Payload schema (pass as JSON string in `payload_json`):
```
{
  "filename_hint": "Optional preferred filename stem",
  "client": "Optional client name for indexing",
  "agent": "Optional agent identifier",
  "retention_days": 14,                       // min 14, max 90
  "blocks": [
    {"type":"heading", "level":1|2|3|4, "text":"..."},
    {"type":"paragraph", "text":"Supports **bold**, *italic*, `code` inline."},
    {"type":"bullets", "items":["...","..."]},
    {"type":"numbered", "items":["...","..."]},
    {"type":"table", "headers":["..."], "rows":[["..."]]},
    {"type":"callout", "severity":"info|warning|critical", "text":"..."},
    {"type":"page_break"}
  ]
}
```
Returns: { id, filename, download_url, expires_at, size_bytes }. The download_url is signed and surfaced in chat as a clickable link.

RCS Document Generator

Model	Context	Best for
GPT-5.5	1.05M	Frontier reasoning for complex professional workloads. Highest reliability on hard problems; most expensive GPT option.
GPT-5.6 Terra	1.05M	Best everyday GPT. Balances intelligence and cost — half the price of the flagship tier with strong general capability.
GPT-5.6 Luna	1.05M	Fast and cheap for high-volume simple work. ⚠️ Noticeably weaker at long-context recall — don't feed it huge documents.
GPT-5.4 Mini	400K	Budget tier. Text + image input, tool use, web/file search. Good for bulk, low-complexity tasks.

Model	Context	Best for
Claude Opus 5	1M	Highest capability. Deep reasoning, long analysis, hardest coding. No long-context price penalty — best choice for very large documents.
Claude Sonnet 5	1M	Daily driver. Most agentic Sonnet yet — sustained coding, multi-step automation, debugging, research. Near-Opus quality at much lower cost.
Claude Haiku 4.5	200K	Fastest and lightest. Quick lookups, short drafts, summaries, high-volume simple tasks.
Claude Opus 4.8	1M	Previous premium tier. Still excellent for accuracy-critical work; superseded by Opus 5 at the same price.
Claude Sonnet 4.6 (1M)	1M	Previous general-purpose default. Superseded by Sonnet 5 — use Sonnet 5 unless you need output consistency with older work.

Model	Context	Best for
Gemini 3.1 Pro	1M	Google flagship. Advanced reasoning, agentic workflows, and coding; strongest multimodal handling (text, image, audio, video, PDF).
Gemini 3.6 Flash	1M	Best everyday Gemini. Fast and cheap with a current knowledge cutoff (March 2026); beats 3.5 Flash on coding and long-context.
Gemini 3.5 Flash	1M	Previous Flash generation. Superseded by 3.6 Flash, which is both cheaper and better.

Login to RCS AI

RCS AI — Available Model Overview

OpenAI (GPT)

Anthropic (Claude)

Google (Gemini)