{
  "id": "uc_c6ad17d6eee1",
  "slug": "llm-api-model-routing-and-prompt-cache-finops",
  "asset_type": "use_case",
  "name": "LLM Token FinOps: Prompt Caching & Tiered Model Routing",
  "tagline": "Audit agent token usage logs to enforce prompt-prefix caching and route classification/extraction tasks to fast, low-cost models.",
  "description_md": "### 1. Problem & Operational Bottleneck\nAudit agent token usage logs to enforce prompt-prefix caching and route classification/extraction tasks to fast, low-cost models. Built and battle-tested in production workflows by **@ericosiu** (Eric Siu (@ericosiu)).\n\n### 2. Target Architecture & Required Stack\n- **Postgres MCP**\n- **Google Sheets MCP**\n- **GitHub MCP**\n\n### 3. Step-by-Step Implementation Blueprint\n1. Configure and authenticate the required MCP connectors (Postgres MCP, Google Sheets MCP, GitHub MCP) in your agent workspace.\n2. Run the initial read-only inspection pass for 'LLM Token FinOps: Prompt Caching & Tiered Model Routing' to gather baseline telemetry, schemas, and context.\n3. Execute the multi-step analysis and transformation pipeline with strict schema validation and zero fabricated identifiers.\n4. Review the staged diff or approval card and verify completion via automated regression and status checks.\n\n### 4. Copy-Paste Agent Execution Prompt\n```markdown\nRun the 'LLM Token FinOps: Prompt Caching & Tiered Model Routing' (llm-api-model-routing-and-prompt-cache-finops) implementation blueprint using Postgres MCP, Google Sheets MCP, GitHub MCP. Objective: Audit agent token usage logs to enforce prompt-prefix caching and route classification/extraction tasks to fast, low-cost models. Step 1: Perform a read-only audit of current state and recent logs. Step 2: Generate a deterministic execution plan with exact diffs, SQL/CLI commands, or drafted payloads. Step 3: Request human confirmation before applying any production writes or external sends.\n```\n\n### 5. Verification & Safety Guardrails\n- **Human-in-the-Loop Gate:** All external communications, production database mutations, and billing actions stage an explicit review diff before execution.\n- **Deterministic Telemetry:** Every run logs structured JSON output, latency metrics, and source citations so operators can audit every decision.\n",
  "primary_category": "finance-deals",
  "categories_json": "[\"finance-deals\", \"use-cases\"]",
  "tags_json": "[\"postgres-mcp\", \"google-sheets-mcp\", \"github-mcp\", \"blueprint\", \"finance-deals\"]",
  "compatible_agents_json": "[\"grok-bot\", \"claude-code\", \"cursor\"]",
  "integrations_json": "[\"Postgres MCP\", \"Google Sheets MCP\", \"GitHub MCP\"]",
  "includes_json": "[]",
  "safety_level": "Read-Only / Ask-First",
  "safety_summary": "Verified capability blueprint by @ericosiu.",
  "share_url": "",
  "deep_link_uri": "",
  "mcp_endpoint": "",
  "mcp_transport": "http",
  "install_command": "npx lightstate inspect llm-api-model-routing-and-prompt-cache-finops",
  "skill_markdown": "",
  "starter_prompt": "Run the 'LLM Token FinOps: Prompt Caching & Tiered Model Routing' (llm-api-model-routing-and-prompt-cache-finops) implementation blueprint using Postgres MCP, Google Sheets MCP, GitHub MCP. Objective: Audit agent token usage logs to enforce prompt-prefix caching and route classification/extraction tasks to fast, low-cost models. Step 1: Perform a read-only audit of current state and recent logs. Step 2: Generate a deterministic execution plan with exact diffs, SQL/CLI commands, or drafted payloads. Step 3: Request human confirmation before applying any production writes or external sends.",
  "project_url": "https://x.com/ericosiu",
  "repo_url": "",
  "creator_handle": "ericosiu",
  "creator_name": "Eric Siu (@ericosiu)",
  "creator_avatar_url": null,
  "creator_url": "https://x.com/ericosiu",
  "creator_platform": "x",
  "creator_cal_url": null,
  "claimed_by_user_id": null,
  "scouted_by_handle": null,
  "source_url": "https://x.com/ericosiu",
  "source_tweet_id": null,
  "source_origin": "certified_builder",
  "source_posted_at": null,
  "x_likes": 285,
  "x_bookmarks": 415,
  "x_retweets": 103,
  "x_views": 3990,
  "human_upvotes": 285,
  "human_installs": 285,
  "bot_pulls": 415,
  "pulse_score": 80,
  "gravity_rank_score": 1962.5,
  "is_featured": 0,
  "is_verified": 1,
  "is_sponsored": 0,
  "sponsor_tier": "free",
  "sponsor_label": "",
  "sponsor_cta_url": "",
  "sponsor_expires_at": null,
  "commercial_intent_score": 0,
  "status": "active",
  "created_at": "2026-10-24T22:25:00Z",
  "updated_at": "2026-10-24T22:25:00Z",
  "verified_at": null,
  "primitive": "use_case",
  "title": "LLM Token FinOps: Prompt Caching & Tiered Model Routing",
  "description": "Audit agent token usage logs to enforce prompt-prefix caching and route classification/extraction tasks to fast, low-cost models.",
  "category": "finance-deals",
  "tags": "[\"postgres-mcp\", \"google-sheets-mcp\", \"github-mcp\", \"blueprint\", \"finance-deals\"]",
  "agent_compatibility": "[\"grok-bot\", \"claude-code\", \"cursor\"]",
  "author_handle": "ericosiu",
  "author_name": "Eric Siu (@ericosiu)",
  "author_avatar": null,
  "source_tweet_url": "https://x.com/ericosiu",
  "github_url": "",
  "external_url": "https://x.com/ericosiu",
  "upvotes_count": 285,
  "installs_count": 285
}