Index of CitizenWeb3 AI agents. This main branch carries only this README; each agent lives on its own long-running branch in this repo, in a sibling repo, or locally.
| Agent | Location | Purpose | Stack | Status |
|---|---|---|---|---|
| Aida — Telegram Growth Agent | this repo, branch aida-trust-first-rewrite |
Autonomous Telegram agent: joins Web3/staking/privacy groups, answers via ValidatorInfo + CW3 podcast RAG, grows community | Python 3.12, Telethon, aiogram, Postgres, Claude Code CLI, Docker | Active |
| AI Tools Landing | this repo, branch ai-landing |
Public landing for CW3 AI tools | Next.js, TypeScript, Framer Motion, Tailwind | Active |
| Logos Chatbot | this repo, branch logos-onboarding-assistant · logos.staking.citizenweb3.com |
RAG chatbot for the Logos network — onboarding assistant for users, developers, and validators | Next.js 16, AI SDK, Drizzle, Postgres, Redis, Vertex AI | Active |
| Logos Node Skill | this repo, branch logos-node |
AI agent skill: install, update, and operate a Logos Blockchain testnet validator node from natural language | Node.js, bash, systemd, npx | Active |
| AI Integrations CLI | this repo, branch installer |
npm CLI (@citizenweb3/ai-integrations) that installs agent skills into Claude Code, GitHub Copilot, Gemini CLI, Codex CLI, opencode, OpenClaw |
Node.js, npm | Active |
| Upwork Agent | citizenweb3/upwork-agent | Autonomous Upwork job search + scoring + proposal drafting + submission, human-in-the-loop via Telegram | Node.js, Chrome DevTools Protocol, Claude Code CLI, Telegram bot, Docker | Active |
| ValidatorInfo — RAG Assistant | citizenweb3/validatorinfo | In-product RAG endpoint over CW3 podcasts + on-chain validator data; serves /api/rag/search consumed by Aida, Content Creator, Bizdev Email Agent |
Next.js API route, Postgres, Prisma | Active |
| ValidatorInfo — Fullstack Developer | citizenweb3/validatorinfo (agents-infrastructure/) |
Self-hosted GitHub Actions runner that ships features/fixes against the ValidatorInfo codebase via Claude Code | Docker, GitHub self-hosted runner, Claude Code CLI, DeepContext MCP, GitNexus MCP, Figma MCP | Active |
| ValidatorInfo — Content Creator | citizenweb3/validatorinfo (agents-infrastructure/) |
Two self-hosted runners (per-brand) generating and publishing content to Twitter / Telegram / Discord from indexed data | Docker, GitHub self-hosted runner, Claude Code CLI, Twitter API, Telegram, Discord webhooks | Active |
| Bizdev Email Agent | local, not yet published | Autonomous outreach: prospect discovery, research, personalized cold drafting, Telegram approval, Resend send, reply capture, self-improvement via embeddings | Next.js dashboard, TypeScript worker, Postgres + pgvector, Drizzle, Resend, Telegram bot, Apollo.io, Hunter.io | WIP (local) |
Autonomous agent for CitizenWeb3. Joins Web3/staking/privacy Telegram groups, answers questions using real on-chain data (ValidatorInfo) + CW3 podcast RAG, posts under human approval.
Architecture: main.py event loop → Telethon listens to groups → trust gates filter messages → Claude CLI generates reply → approval bot (aiogram) routes to operator → on approval, posts back. Postgres stores history; healthcheck + autoheal keep container alive.
Branches:
| Branch | Role | Last activity |
|---|---|---|
aida-trust-first-rewrite |
Canonical. Trust-first rewrite — always-Phase-2 trust gates, post-Phase-2 hard gate, hostility/defensiveness handling | latest |
tg-growth-agent |
Original mainline before rewrite | 3 weeks ago |
tg-growth-agent-llama |
Experimental: Ollama Qwen3-30B-A3B as pre-filter to cut Claude calls | 3 weeks ago |
prompt-extraction-refactor |
Local-only refactor branch (not pushed) — prompt extraction, drops already_offered tracking |
2 weeks ago |
Public-facing landing page presenting CitizenWeb3 AI tools/agents.
Architecture: single-page Next.js App Router (src/app/page.tsx) with Framer Motion animations, Tailwind v4 styling, Lucide icons. No backend, no auth, no DB.
Branches:
| Branch | Role | Last activity |
|---|---|---|
ai-landing |
Canonical. Original landing build | 6 weeks ago |
ai-landing-dev |
Design + copy refresh, not yet merged back | 3 weeks ago |
Live: logos.staking.citizenweb3.com · Branch: logos-onboarding-assistant
RAG chatbot for the Logos network. Onboarding assistant for three audiences: end users (network basics, getting started), developers (building on Logos), and validators (running infrastructure). Answers grounded in indexed Logos documentation and ecosystem sources.
Architecture: Next.js 16 App Router + AI SDK v6 (@ai-sdk/google-vertex, Gemini provider). Drizzle ORM over Postgres (pgvector) for chat logs, sources, chunks. Redis (ioredis) for caching and rate limiting. Background indexer/ process ingests sources → chunks → embeddings on a cron schedule. Services in src/app/services/: chat-log-service, chunk-service, rate-limit-service, source-service, retrieval-service, rerank-service. Fully containerised via Docker Compose (app, indexer, postgres, redis).
Data sources indexed: logos.co, build.logos.co, docs.waku.org, press.logos.co, blog.nomos.tech; GitHub orgs logos-co and logos-blockchain; curated raw GitHub docs (logos-docs, logos-lips, logos-blockchain-specs).
Branch: logos-node
AI agent skill that lets you install, update, and operate a Logos Blockchain testnet validator node using natural language inside any supported AI agent.
Commands:
| Command | Action |
|---|---|
/logos-node install |
Fresh node installation: binary, circuits, systemd service, faucet |
/logos-node update |
Breaking upgrade: wipe state, re-init, restart with new binary |
/logos-node status |
Check sync mode, peers, block height, wallet balance |
Install:
npx @citizenweb3/ai-integrations logos-nodeRequirements: Linux x86_64, glibc ≥ 2.39 (Ubuntu 24.04+ / Debian 12+), ≥ 64 GB free storage, root/sudo.
Branch: installer · npm: @citizenweb3/ai-integrations
npm package + CLI that installs Citizen Web3 agent skills into any supported AI coding agent. Auto-detects installed agents and asks before writing to each.
Supported agents:
| Agent | Skill path |
|---|---|
| Claude Code | ~/.claude/skills/<network>/ |
| GitHub Copilot | .agents/skills/<network>/ |
| Gemini CLI | .gemini/skills/<network>/ |
| OpenAI Codex CLI | .agents/skills/<network>/ |
| opencode | .opencode/skills/<network>/ |
| OpenClaw | ~/.openclaw/skills/<network>/ |
Available skills:
| Skill | Status |
|---|---|
logos-node |
✅ Available |
darkfi-node |
🔜 Planned |
genlayer-node |
🔜 Planned |
Repo: citizenweb3/upwork-agent
Autonomous Upwork job-search and proposal agent powered by Claude Code. Background daemon ties three components together: a logged-in Google Chrome session (via Chrome DevTools Protocol), a Telegram bot for human-in-the-loop control, and Claude Code as the browsing/scoring/writing brain.
Architecture: cron tick (or /search) → Claude Code drives Chrome, scrapes job listings → scores each job 0-10 against operator's profile → jobs ≥ 6 sent to Telegram as cards with inline buttons (Apply / Skip) → on Apply, Claude generates a proposal matching operator's writing style → operator reviews (Send / Cancel / Redo) → on Send, Claude submits via the same Chrome session. Two run modes: Docker (server, headless, Chrome via noVNC at :6080) and bare-metal (Mac/Linux desktop with native Chrome).
Repo: citizenweb3/validatorinfo
ValidatorInfo is a Web3 multi-chain explorer (Cosmos, Polkadot, Ethereum, Namada, Aztec, …). Three-tier system: Next.js frontend (Socket.IO + next-intl), indexer service (background cron jobs), Postgres + Prisma + Redis. Three AI agents live alongside the product — one in-app (RAG Assistant) and two as self-hosted GitHub Actions runners (agents-infrastructure/).
In-product RAG endpoint at src/app/api/rag/search/. Indexes the CitizenWeb3 podcast corpus + on-chain validator data; serves chat responses inside ValidatorInfo and exposes RAG_API_URL + RAG_API_TOKEN for external clients (Aida, Bizdev Email Agent, Content Creator). Single internal API surface — /api/rag/search — consumed across the agent fleet.
Self-hosted GitHub Actions runner (label self-hosted,builder) that picks up workflow jobs against the ValidatorInfo repo and ships features/fixes via Claude Code. Wired with full project context: DATABASE_URL + EVENTS_DATABASE_URL (read access to validator + events DBs), WILDCARD_API_KEY (DeepContext semantic code search), GitNexus for code-relationship/impact analysis, FIGMA_ACCESS_TOKEN for design pulls, GH_TOKEN for PR operations. Persistent volumes hold ~/.claude state, runner state, workspace, and GitNexus cache across runs.
Two parallel content-publishing runners — one per brand:
agent-content→ ValidatorInfo (@therealvalinfo)agent-content-cw3→ CitizenWeb3 (@citizen_web3)
Each is a self-hosted runner triggered by GitHub workflows, generating content from indexed data and publishing to Twitter (OAuth1 + OAuth2), Telegram channel, and Discord webhook. Pulls source material via the RAG Assistant (RAG_API_URL) and the validator DB. Helper scripts in agents-tools/: post-twitter.js, post-telegram.js, post-discord.js, search-rag.js, log-social.sh. Brand isolation enforced via separate Twitter/Telegram/Discord credentials and separate runner workspaces.
Status: WIP, local-only. No public repo yet. MVP scaffold in ~/project/dev/bizdev-email-agent.
Autonomous outbound campaign agent for CitizenWeb3. End-to-end pipeline: prospect discovery → research/enrichment → personalized drafting → operator approval → send → reply capture → tracking → self-improvement via email embeddings.
Architecture: Next.js operator dashboard (SSR) + TypeScript worker with Postgres-backed leasing (commands / jobs / job_runs / event_log / work_items) + Postgres + pgvector + Drizzle. Cascading prospect discovery (webread → GitHub → Apollo.io → Hunter.io) with confidence protocol (≥ 80 % to draft). Sonnet first draft + Opus on redo. Telegram approval (aiogram) — batch + single review. Resend for outbound on partner@citizenweb3.com + Resend Inbound webhook + Cloudflare Email Routing for replies. ValidatorInfo Postgres (read-only) + RAG API (/api/rag/search) as data sources. Email embeddings (sqlite-vec / embeddinggemma-300m) for self-improving drafts.
Two pillars:
- Staking Growth — VCs, foundations, DAOs, HNW stakers → migrate delegations to CW3 bare-metal validators.
- AI Workforce — projects, explorers, wallets, DAOs ($1M+ TVL) → sell autonomous AI agent architecture (ValForge + specialized agents), ValidatorInfo as live flagship.
Multi-touch model:
- Cold — first-touch after research + confidence check.
- Warm — agent drafts replies to inbound responses; operator approves via Telegram.
- Direct — operator's manual replies routed through Telegram bot → Resend, thread stays on
partner@citizenweb3.com. - Roadmap: follow-up sequences (5/12/20-day), warm-reply automation, multi-channel (Twitter/TG/Discord/LinkedIn DMs).
MIT © Citizen Web3
- Org: https://github.com/citizenweb3
- This repo: https://github.com/citizenweb3/ai-integrations
- ValidatorInfo: https://github.com/citizenweb3/validatorinfo
- Upwork Agent: https://github.com/citizenweb3/upwork-agent