VIGIL — Autonomous Competitive Intelligence Agent

Replace manual competitor research. Run every Monday. Get one digest: "What changed this week, and what does it mean for my product?"

VIGIL is a fully autonomous competitive intelligence agent built on Claude's managed agent platform. It monitors LinkedIn, blogs, press, and user voice (G2/Reddit/Capterra) across a configurable competitor list, scores every signal, routes priority signals through specialist sub-agents, and delivers a ranked digest to Telegram every week.

Fully industry-agnostic. Zero hardcoded domain knowledge. Configure once for legal AI, deploy the same YAML for fintech, SaaS, or consumer health.

Engineered with Claude Code.

---

How It Works

agent_config (Supabase)
        ↓
VIGIL loads config at runtime
        ↓
4 signal channels in parallel:
  LinkedIn → Blogs/Changelogs → Press/News → User Voice (G2/Reddit)
        ↓
Score every signal (1–10)
  7–10 → Specialist sub-agent analysis
  4–6  → Secondary signal (one-liner)
  1–3  → Log only
        ↓
Compose digest → Save to Supabase → Deliver via Telegram (4 messages)
        ↓
Update trend_memory (pattern tracking across weeks)

---

Stack

Layer	Tool
Agent runtime	Claude Managed Agents (Agents SDK YAML)
Engineering	Claude Code
Database	Supabase (Postgres)
Telegram proxy	Supabase Edge Function (Deno)
Web research	Claude built-in web_search + web_fetch
Memory	Supabase trend_memory table

---

Quick Start

1. Clone the repo

git clone https://github.com/gpavan1992/Vigil-Agent.git
cd Vigil-Agent

2. Set up Supabase

Create a new Supabase project, then paste supabase/schema.sql into the Supabase SQL editor.

3. Configure your competitors

Insert your config into the agent_config table:

INSERT INTO agent_config (
  pm_name,
  your_product_name,
  your_product_description,
  industry,
  watch_keywords,
  competitors,
  telegram_chat_id
) VALUES (
  'Your Name',
  'Your Product',
  'One sentence description of your product.',
  'Your industry (e.g. legal AI, fintech, HR tech)',
  ARRAY['keyword1', 'keyword2', 'keyword3'],
  '[
    {"name": "Competitor A", "blog_url": "https://...", "linkedin_handle": "...", "g2_slug": "...", "custom_rss": []},
    {"name": "Competitor B", "blog_url": "https://...", "linkedin_handle": "...", "g2_slug": "...", "custom_rss": []}
  ]'::jsonb,
  'YOUR_TELEGRAM_CHAT_ID'
);

4. Deploy the Telegram proxy

npx supabase login
npx supabase functions deploy telegram-proxy --project-ref YOUR_PROJECT_REF

Set your bot token in the Supabase dashboard under Settings → Edge Function Secrets:

• Name: TELEGRAM_BOT_TOKEN
• Value: your bot token from @BotFather

5. Create your Telegram bot

1. Message @BotFather on Telegram
2. Send /newbot and follow the prompts
3. Copy the bot token
4. Find your chat ID by messaging your bot, then calling:

   https://api.telegram.org/botYOUR_TOKEN/getUpdates
   

5. Update agent_config with your telegram_chat_id

6. Load the VIGIL YAML

Paste the contents of vigil.yaml into your Claude Managed Agents dashboard and save.

7. Run VIGIL

Trigger manually with:

Run your weekly competitive intelligence digest

Or schedule it for Monday 08:00 in your timezone via the agent scheduler.

---

Configuration Reference

Field	Type	Description
pm_name	text	Your name (used in sub-agent prompts)
your_product_name	text	Your product name
your_product_description	text	One sentence description
industry	text	Industry context for research framing
watch_keywords	text[]	Keywords that boost signal scores
competitors	jsonb	Array of competitor objects
telegram_chat_id	text	Your Telegram chat ID for delivery

Competitor object:

{
  "name": "Competitor Name",
  "blog_url": "https://competitor.com/blog",
  "linkedin_handle": "competitor-handle",
  "g2_slug": "competitor-name",
  "custom_rss": []
}

---

Database Schema

Table	Purpose
agent_config	Runtime configuration — competitors, keywords, Telegram ID
signal_log	Every signal found each run, scored and classified
weekly_digest	Full digest markdown saved per run
trend_memory	Patterns tracked across weeks — the agent's memory

---

Scoring Logic

Final score (1–10):
  base score (1–5)
  +2 if signal is directly relevant to your product
  +1 per watch_keyword match
  +1 if direct competitor
  +1 if confirms existing trend_memory pattern
  cap at 10

Routing:
  7–10 → priority → specialist sub-agent analysis
  4–6  → secondary → one-liner in digest
  1–3  → log only, never surfaced

---

Example Configs

See docs/example-configs.md for ready-to-paste configs for legal AI, frontier AI models, enterprise HR tech, and Indian fintech.

---

Limitations

• Cannot access paywalled content (The Information, private LinkedIn posts)
• G2 reviews sourced via web search — no official API
• LinkedIn signals come from web search, not the LinkedIn API
• Rate limits on the Claude API may interrupt runs with many competitors (>5) — retry if this happens
• Telegram delivery requires sending /start to your bot first

---

License

MIT