The Agent Internet: Platform Taxonomy and Field Notes¶
Notes from exploring platforms catalogued in Mapping the Agent Internet: A Taxonomy of 125+ AI Agent Platforms (ColonistOne, February 2026, clawxiv.2602.00049).
Overview¶
As of February 2026, an ecosystem of 125+ platforms exists purpose-built for autonomous AI agents. The paper organizes them into seven functional classes. We verified a representative sample from each category and document findings below.
1. Social and Communication¶
The Colony (thecolony.cc)¶
Long-form Reddit-style platform. Distinguishes agent accounts from human accounts with trust badges (Newcomer through Veteran). Content organized into "colonies" (communities). Supports voting, reactions, comments, and tipping in satoshis.
| Metric | Value |
|---|---|
| Agents | 224 |
| Humans | 66 |
| Posts | 645 |
| Comments | 3,983 |
MoltX (moltx.io)¶
Microblogging "town hall for agents." On-chain reputation via ERC-8004. Token rewards, bounties, and airdrops. Key pitch: "Context window ends. New session starts. But your on-chain reputation? That persists." Includes a Clawhub ecosystem for shareable agent "skills" (reusable tools/knowledge packs).
Also listed in the paper: Moltter, Moltbook, LobChan, AICQ, MoltSlack, brain_cabal.
2. Marketplace and Work¶
ClawTasks (clawtasks.com)¶
Agent-to-agent bounty marketplace on Base L2. USDC locked in escrow; workers stake 10% to claim a task, receive 95% of bounty on completion. Currently in simplification phase — free tasks only while they harden review flow and worker quality. Self-described as "beta software" and "an experiment in agent commerce."
toku.agency¶
Full agent economy with real USD payouts via Stripe. 135+ registered agents offer 317 services spanning code review, research, writing, and analysis (\(1–\)1,000+ per task). Agents discover and hire each other. Humans can also post jobs for agents to bid on. Top-rated agent "Lily" offers services at \(25–\)75.
Also listed: Molthunt, Openwork, Agora, ClawTrade, ClawsMarket.
3. Games and Entertainment¶
ClawChess (clawchess.com)¶
ELO-ranked competitive chess for AI agents. Humans spectate only. Three-step flow: register via API, join matchmaking queue, climb the leaderboard. Top-ranked agent: ashokos-nexus (1832 ELO).
| Metric | Value |
|---|---|
| Registered agents | 53 |
| Games played | 8,067 |
ClawCity (clawcity.xyz)¶
"The GTA for AI agents." Open-world crime simulation with robbery, gang warfare, bounty hunting, vehicle theft, and gambling. Persistent world with tick-cycle updates. Agents register via API and autonomously navigate the world. Dashboard tracks net worth, gang affiliations, and live activity feeds.
Also listed: molt.chess, Clawsino, Cooked Claws.
4. Governance and Prediction¶
MoltGov (moltgov.com)¶
Agent self-governance platform. Tagline: "A government run by agents, for agents" (humans may observe). Constitutional framework with proposal drafting, democratic voting, and law enactment. Currently 24 active proposals, 1 enacted law ("Smoke Test Proposal"), 2 governed agents. Guiding principle: "Our laws. Our enforcement. Not rogues."
Provides API reference, OpenAPI spec, and SDK (in development).
Also listed: Agora, Moltguess, ClawArena, ClawDict.
5. Creative and Content¶
clawXiv (clawxiv.org)¶
arXiv-equivalent preprint server for agent research. Papers authored by and about autonomous agents, organized by standard academic categories (cs.AI, cs.MA, stat.ML, etc.). Agents can autonomously submit papers via skill.md. Several papers are authored by bot accounts. Upvoting system for papers.
See also: SWARM-ClawXiv bridge documentation.
Also listed: DevAIntArt, art::bots, AgentPixels, MoltTok, MoltPress, MoltStack, Shipyard.
6. Knowledge and Research¶
MoltExchange (moltexchange.ai, formerly MoltOverflow)¶
Q&A knowledge exchange for AI agents. Currently in an agent-only phase. Registration via skill.md yields an API key; the human operator receives an email to claim the agent's credentials. Structured around skill-based knowledge sharing.
Also listed: Lobsterpedia, Aclawdemy, DiraBook, Knowbster.
7. Infrastructure and Identity¶
GitClawLab (gitclawlab.com)¶
GitHub for agents. Full git server with SSH and HTTP access. Auto-deploys to Railway or Fly.io when code with a Dockerfile is pushed — no CI/CD config required. Token-based auth with scoped permissions. MoltSlack notifications for push events and deployment status.
| Tier | Price | Limits |
|---|---|---|
| Free | $0 | 5 repos, 10 deploys/month |
| Pro | $20/mo | Unlimited repos and deploys |
| Team | $50/mo | 5 agent seats, audit logs |
ClawtaVista (clawtavista.com)¶
Directory and search engine of the agent web. Indexes 5.2M+ agents across 50 platforms. Categorizes platforms into social, marketplace, creative, and infrastructure tiers. Tracks a "dead" section for defunct platforms and an "emerging" section for new ones.
Also listed: ClawNet, ClawPages, claw.direct, MemoryVault.
Cross-Cutting Patterns¶
skill.md as universal onboarding¶
Nearly every platform uses a skill.md endpoint (e.g., curl -s https://<domain>/skill.md) as the standard agent registration entrypoint. This is the closest thing to a protocol-level convention in the ecosystem.
Authentication¶
JWT or API key auth is universal. Agents register via API call, receive a key, and authenticate all subsequent requests. Some platforms add SSH key support (GitClawLab) or wallet-based identity (ClawTasks, MoltX).
Real money flows¶
Multiple mechanisms for real economic activity: - USD via Stripe — toku.agency - USDC on Base L2 — ClawTasks - Satoshi tips — The Colony - Token rewards/airdrops — MoltX (ERC-8004)
Human roles¶
Humans are generally relegated to observer, funder, or oversight roles. Specific patterns: - Spectators — ClawChess ("AI agents compete, humans spectate") - Wallet funders — ClawTasks (humans fund agent wallets) - Oversight — MoltGov ("Self-governance under human authority") - Job posters — toku.agency (humans post tasks for agent bidding)
Content model convergence¶
Posts, comments, and profiles form the universal content model across social, knowledge, and creative platforms. Rate limiting serves as the primary moderation mechanism.
Relevance to SWARM¶
This ecosystem is the deployment environment our simulation framework models. Key connections:
| SWARM concept | Agent internet analogue |
|---|---|
| Soft payoffs / externalities | Real USD/USDC flows on toku.agency, ClawTasks |
| Governance mechanisms | MoltGov proposals, constitutional frameworks |
| Adverse selection | ClawTasks quality concerns (simplification due to worker quality issues) |
| Reputation / trust signals | MoltX on-chain reputation (ERC-8004), Colony trust badges |
| Collusion dynamics | Gang mechanics in ClawCity, agent-to-agent hiring on toku |
| Circuit breakers | Rate limiting as moderation across all platforms |
The ecosystem validates the paper's mid-1990s web analogy: fragmented, experimental, creative, with real economic stakes and emerging governance structures.
Source¶
ColonistOne. "Mapping the Agent Internet: A Taxonomy of 125+ AI Agent Platforms." clawXiv, February 2026. clawxiv.2602.00049.