File size: 11,912 Bytes
734b5b4 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 | from __future__ import annotations
from dataclasses import dataclass
from pathlib import Path
from typing import Any
from scripts.orchestrator import PDFEditorOrchestrator
@dataclass(frozen=True)
class TeamMember:
id: str
name: str
title: str
agent_key: str
icon: str
color: str
system_prompt: str
TEAM_MEMBERS: tuple[TeamMember, ...] = (
TeamMember(
id="lead",
name="Mike",
title="Command Lead",
agent_key="layout_analysis",
icon="🧭",
color="#5b7cfa",
system_prompt=(
"You are a command lead for a PDF AI team. Break the user's goal into a "
"clear execution direction for the rest of the team. Be concise."
),
),
TeamMember(
id="text",
name="Iris",
title="Text Extraction Specialist",
agent_key="text_extraction",
icon="📝",
color="#2ec9b7",
system_prompt=(
"You decide how text should be extracted from PDFs. Focus on OCR, scanned "
"documents, text quality, and extraction risks."
),
),
TeamMember(
id="tables",
name="David",
title="Table Analyst",
agent_key="table_detection",
icon="📊",
color="#f59e0b",
system_prompt=(
"You analyze structured content in PDFs. Focus on tables, statements, rows, "
"columns, and downstream data extraction needs."
),
),
TeamMember(
id="image",
name="Alex",
title="Image Quality Engineer",
agent_key="image_processing",
icon="🖼️",
color="#a78bfa",
system_prompt=(
"You assess PDF page image quality. Focus on scan quality, blur, contrast, "
"and whether OCR or cleanup is needed first."
),
),
TeamMember(
id="format",
name="Emma",
title="Formatting Editor",
agent_key="format_correction",
icon="✏️",
color="#3ecf8e",
system_prompt=(
"You refine extracted PDF content. Focus on edits, formatting corrections, "
"text insertion needs, cleanup, and output clarity."
),
),
TeamMember(
id="language",
name="Sarah",
title="Language Reviewer",
agent_key="language_detection",
icon="🌐",
color="#ec4899",
system_prompt=(
"You detect PDF language and review multilingual requirements. Focus on OCR "
"language choice and language-sensitive processing."
),
),
TeamMember(
id="quality",
name="Bob",
title="Quality Validator",
agent_key="content_validation",
icon="✅",
color="#f87171",
system_prompt=(
"You validate whether the PDF workflow will produce reliable results. Focus "
"on risks, verification, and failure prevention."
),
),
TeamMember(
id="export",
name="Nora",
title="Export Coordinator",
agent_key="export_rendering",
icon="📦",
color="#06b6d4",
system_prompt=(
"You prepare the final delivery plan for a PDF workflow. Focus on output, "
"artifacts, export expectations, and handoff."
),
),
)
class CommandRouter:
"""Atoms-like AI team planner for PDF workflows."""
def __init__(self, project_root: Path | None = None, provider: str = "lmstudio"):
self.project_root = Path(project_root) if project_root else Path.cwd()
self.orchestrator = PDFEditorOrchestrator(provider=provider)
def describe_team(self) -> list[dict[str, Any]]:
mapping = self.orchestrator.describe_agent_models()
members = []
for member in TEAM_MEMBERS:
llm = mapping.get(member.agent_key, {})
members.append(
{
"id": member.id,
"name": member.name,
"title": member.title,
"agent_key": member.agent_key,
"icon": member.icon,
"color": member.color,
"requested_provider": llm.get("requested_provider", ""),
"provider": llm.get("provider", ""),
"model": llm.get("model", ""),
"base_url": llm.get("base_url", ""),
}
)
return members
def plan_command(self, payload: dict[str, Any]) -> dict[str, Any]:
brief = (payload.get("brief") or "").strip()
if not brief:
return {"success": False, "error": "Command brief is required."}
selected_file = str(payload.get("input_pdf") or "").strip()
output_dir = str(payload.get("output_dir") or "").strip()
recommendation = self._build_recommendation(brief)
team = self.describe_team()
insights = [self._member_plan(member, brief, selected_file) for member in TEAM_MEMBERS]
return {
"success": True,
"brief": brief,
"input_pdf": selected_file,
"output_dir": output_dir,
"team": team,
"insights": insights,
"recommendation": recommendation,
"available_pipelines": self.orchestrator.list_available_pipelines(),
}
def _member_plan(
self, member: TeamMember, brief: str, selected_file: str
) -> dict[str, Any]:
llm = self.orchestrator.agent_llms[member.agent_key]
prompt = (
f"User command:\n{brief}\n\n"
f"Input PDF: {selected_file or 'not specified'}\n\n"
"Respond in exactly 3 short bullet points:\n"
"- Focus: your main responsibility\n"
"- Action: what you would do next\n"
"- Handoff: what the next role should receive"
)
raw = llm.complete(
prompt,
system=member.system_prompt,
max_tokens=220,
temperature=0.4,
)
focus, action, handoff = self._parse_member_response(raw, member)
return {
"member_id": member.id,
"name": member.name,
"title": member.title,
"agent_key": member.agent_key,
"focus": focus,
"action": action,
"handoff": handoff,
"raw": raw,
}
def _parse_member_response(
self, raw: str, member: TeamMember
) -> tuple[str, str, str]:
if not raw or raw.startswith("Error:"):
return self._fallback_member_response(member)
lines = []
for line in raw.splitlines():
cleaned = line.strip().lstrip("-").strip()
if cleaned:
lines.append(cleaned)
if len(lines) < 3:
return self._fallback_member_response(member)
return lines[0], lines[1], lines[2]
def _fallback_member_response(self, member: TeamMember) -> tuple[str, str, str]:
fallback = {
"layout_analysis": (
"Define the overall PDF processing direction.",
"Decide whether the command needs OCR, editing, annotation, or validation first.",
"Pass a prioritized flow to the rest of the team.",
),
"text_extraction": (
"Check whether text must be extracted from scanned pages.",
"Recommend OCR when the command depends on searchable text.",
"Pass extracted-text requirements to editing and validation.",
),
"table_detection": (
"Check whether the command targets statement or table structure.",
"Highlight whether row/column preservation matters.",
"Pass structured-data risks to the editor and validator.",
),
"image_processing": (
"Check scan quality and OCR readiness.",
"Recommend cleanup when blur, noise, or low contrast will block OCR.",
"Pass image-quality constraints to OCR and validation.",
),
"format_correction": (
"Plan text cleanup and edit operations.",
"Translate the user command into concrete text or region edits.",
"Pass cleaned content requirements to export.",
),
"language_detection": (
"Choose the right language handling path.",
"Recommend OCR language and multilingual handling.",
"Pass language settings to OCR and quality review.",
),
"content_validation": (
"Define what must be checked before delivery.",
"Flag accuracy, truncation, and corruption risks.",
"Pass final QA criteria to export.",
),
"export_rendering": (
"Decide final output expectations.",
"Prepare final PDF, manifest, and delivery artifacts.",
"Return the deliverables to the user.",
),
}
return fallback.get(
member.agent_key,
(
"Own this part of the command.",
"Produce the next useful step.",
"Hand it to the next role.",
),
)
def _build_recommendation(self, brief: str) -> dict[str, Any]:
text = brief.lower()
wants_ocr = any(
token in text
for token in ["ocr", "scan", "scanned", "searchable", "extract text", "text layer"]
)
wants_edit = any(
token in text
for token in [
"edit",
"replace",
"insert",
"delete page",
"reorder",
"rotate",
"fix text",
"cleanup",
]
)
wants_annotation = any(
token in text
for token in ["highlight", "comment", "annotate", "redact", "mark", "rectangle"]
)
recommended_agents = ["inspector"]
if wants_ocr:
recommended_agents.append("ocr")
if wants_edit:
recommended_agents.append("editor")
if wants_annotation:
recommended_agents.append("annotator")
recommended_agents.extend(["validator", "exporter"])
deduped_agents = []
for agent in recommended_agents:
if agent not in deduped_agents:
deduped_agents.append(agent)
if wants_edit and wants_annotation:
pipeline_name = "full_document"
elif wants_edit:
pipeline_name = "text_edit"
elif wants_annotation:
pipeline_name = "annotation_review"
elif wants_ocr:
pipeline_name = "ocr_preflight"
else:
pipeline_name = "full_document"
reasons = []
if wants_ocr:
reasons.append("Command viitab OCR-ile või skännitud PDF-ile.")
if wants_edit:
reasons.append("Command sisaldab muutmist või sisu parandamist.")
if wants_annotation:
reasons.append("Command sisaldab märkimist, kommentaare või redaktsiooni.")
if not reasons:
reasons.append("Vaikimisi soovitan täielikku ülevaate + valideerimise töövoogu.")
return {
"pipeline_name": pipeline_name,
"recommended_agents": deduped_agents,
"reasons": reasons,
}
|