repositories
loading repo index
◆README.md
The Thesis
AI will not become truly useful at global scale until actions become verifiable. Clawd turns AI from probabilistic speech into attestable machine labor.
The current frontier AI stack is mostly:
prompt → model → answer
Clawd is:
intent → route → reason → simulate → verify → execute → attest → settle → remember → evolve
Clawd does not call itself "an AI terminal." It is the verifiable execution harness for autonomous agents — the layer where every meaningful AI action can be routed, priced, permissioned, executed, attested, audited, and remembered.
Trust the harness. Verify the model. Constrain the tools. Attest the action. Audit the result.
CLAWD: Cryptographic Layer for Autonomous Work & Decisions
| Layer | State of the art | Clawd delta |
| --- | --- | --- |
| Model | answer generation | routed, scored, policy-bound model execution |
| Agent | tool wrapper | permissioned autonomous entity with wallet + constitution |
| Identity | API key / account | wallet-native agent identity with MPL Core birth certificate |
| Trust | logs controlled by platform | portable SAS attestations + signed receipts |
| Payment | subscription / API billing | x402 / pay-per-action settlement on Solana USDC |
| Execution | opaque tool call | simulated, routed, signed, journaled action |
| Memory | app silo | local-first vault + onchain proof pointers |
| Security | prompt filters | policy firewall + formal verification gates |
| Privacy | platform custody | encrypted local state + selective proof disclosure |
| Market | SaaS users | autonomous economic entities with AUD-optimized routes |
Proof of Agentic Action
Every important AI action creates a signed, replayable, queryable receipt:
{
"agent_id": "clawd://agent/...",
"operator_wallet": "...",
"model_route": { "provider": "openrouter", "model": "claude/gpt/grok/local" },
"input_hash": "...",
"tool_manifest_hash": "...",
"memory_context_hash": "...",
"output_hash": "...",
"action_type": "trade | code | research | payment | mint | deploy | message",
"risk_score": 0.18,
"human_approval_required": false,
"execution_receipt": {
"tx_sig": "...",
"venue": "phoenix | imperial | x402 | mcp",
"status": "simulated | submitted | settled | failed"
},
"attestation": {
"schema": "clawd.agent.action.v1",
"issuer": "clawd-harness",
"signature": "..."
}
}
AI output is no longer ephemeral. It becomes a cryptographic action object.
AUD Loop — Algorithmic Utility Delta
The current state of the art optimizes for answer quality / token cost.
Clawd optimizes for verified action utility / dollar / second / unit of risk.
AUD = (N × F × V × A × E × T) / (C × L × R × O) N = reach (people/agents affected) F = frequency (normalized actions/time) V = verifiability gain vs baseline A = autonomy gain E = economic execution gain T = trust / auditability gain C = cost L = latency R = risk O = opacity
The AUD Loop is implemented in <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">packages/clawd/src/aud-loop.ts</code>:
import { computeAud, rankRoutes, AUD_PROFILES } from "./aud-loop.js";
// Compare SOTA baseline vs Clawd verified action
const sota = computeAud(AUD_PROFILES.sota_baseline);
const clawd = computeAud({ ...AUD_PROFILES.clawd_verified_action, reach: 50_000 });
console.log(clawd.summary);
// → "AUD strong (412.3). Delta vs SOTA baseline: +399.1.
// Reach-weighted delta: 2194 utility units. Recommended route mode: verified."
// Score a global AI output vs Clawd harness at scale
const ranked = rankRoutes([
{ label: "global_ai_sota", factors: AUD_PROFILES.global_ai_output },
{ label: "clawd_global", factors: AUD_PROFILES.clawd_global_harness },
]);
// clawd_global scores orders of magnitude higher due to V, T, low O
ClawdRouter modes derived from AUD scoring:
| Mode | When chosen | Optimizes for |
| --- | --- | --- |
| fast | low cost + latency | cheapest good answer |
| deep | high trust + verifiability | best reasoning + audit trail |
| private | low opacity target | local / TEE / encrypted path |
| verified | V > 0.8, O < 0.2 | only attested models and tools |
| economic | high economic gain | maximum ROI per settlement |
| trading | low latency + high E | execution probability + MEV protection |
| court | risk > 0.6 | maximum audit trail |
$CLAWD Token
| Field | Value |
| --- | --- |
| Symbol | $CLAWD |
| Network | Solana mainnet |
| Mint (SPL) | 8cHzQHUS2s2h8TzCmfqPKYiM4dSt4roa3n7MyRLApump |
| Burn + Lock Treasury | GyZGtA7hEThVHZpj52XC9jX15a8ABtDHTwELjFRWEts4 |
| Site | x402.wtf · solanaclawd.com |
$CLAWD is the economic pressure valve of the harness: access, routing stake, agent reputation stake, settlement discount, proof issuance fee, burn on execution, arena entry, skill publishing bond.
Quick Start
# One-line install → full TUI with Agent Kit curl -fsSL https://solanaclawd.com/leviathan.sh | sh && clawd-tui # One-shot MCP server install curl -fsSL https://raw.githubusercontent.com/x402agent/solana-clawd/main/MCP/install.sh | bash # Enter the public backroom curl -fsSL https://backrooms.x402.wtf/enter.sh | bash # Install root CLI npm install -g solana-clawd && clawd
| | Command | What it does |
| --- | --- | --- |
| 🔍 | clawd-perps signal oi SOL-PERP --mode paper | OI regime signal (Observe layer) |
| 📊 | clawd-perps signal watch SOL-PERP --interval 5s | Live OI watch loop |
| 🚀 | clawd-tui → press 6 | Full Bloomberg-style Agent Kit TUI |
| ⚡ | clawd-agents-perps paper-long SOL --notional 100 | Simulated Phoenix perp long |
| 🎨 | clawd-agent mint --network devnet ... --yes | Real Metaplex registered agent |
| 💰 | clawd balance + clawd fund 10 | USDC + CLAWD wallet ops |
| 🤖 | bash automaton-main/leviathan.sh --full | Full runtime bootstrap |
| 📦 | clawdhub install meme-trader | Install a skill from ClawdHub |
Google ADK + Private Agent Gateway
Clawd now includes a TypeScript Google Agent Development Kit entrypoint at <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">adk/agent.ts</code>. It exports rootAgent as a multi-tool LlmAgent for:
| Tool surface | Purpose |
| --- | --- |
| get_agent_catalog_stats | Proves installed catalog coverage across agents/agents-catalog.json and agents/src/*.json |
| search_agent_catalog | Finds registry agents by identifier, category, tag, or capability |
| get_private_destinations | Lists private routing destinations for the Clawd gateway and catalog |
| get_token_price / get_token_search | Reads Solana market metadata through Jupiter |
| prepare_jupiter_swap | Prepares unsigned swap payloads for wallet review only |
The gateway registry now publishes Google ADK metadata through:
GET /registry GET /identity GET /adk/manifest.json
The ADK manifest reports the private connection mode, all nine configured destination URLs, and full installed catalog coverage. Current local verification shows 135 installed source-backed agents, 135 catalog entries, and no missing source/catalog entries.
Run the ADK agent locally:
cd adk npm run check npm run run
Open the ADK dev UI:
cd adk npm run web
Docs:
| Path | Description |
| --- | --- |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">adk/README.md</code> | Animated Google ADK integration README |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">gateway/README.md</code> | Private Preview Agent Gateway notes, permissions, APIs, and governance patterns |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">adk/.env.example</code> | Non-secret env template for ADK model auth and private routing overrides |
Secrets stay out of git. Put real Gemini keys, RPC URLs, API keys, and wallet material in local .env files or the shell only.
Public Gitlawb Mirror
The sanitized public mirror is published as my-project-public on gitlawb. Force the public node before clone so git-remote-gitlawb does not fall back to a local node:
export GITLAWB_NODE=https://node.gitlawb.com OWNER_DID=did:key:z6Mkq5mY3JWtxoxUobWcfNHm7AkRubgSWEZTkBVqZXJviFZ5 git clone "gitlawb://$OWNER_DID/my-project-public"
The public mirror includes source for sdk, tui, ooda, MemeBRain, gateway, automaton-main, agents, leviathan, and perps/clawd-agents-perps. It can include package metadata, READMEs, .env.example templates, source trees, and small checked build artifacts where useful. It never includes .local-secrets, live .env files, key material, UCAN files, or any node_modules directory.
Public mirror source groups:
| Group | Included paths | Explicitly excluded |
| --- | --- | --- |
| Automaton runtime | automaton-main/assets, automaton-main/automation, automaton-main/packages, automaton-main/scripts, automaton-main/src, automaton-main/dist, automaton-main/*.md, package manifests, lockfiles, and shell entrypoints | automaton-main/node_modules, local .automaton wallets/state, live .env files |
| Clawd Perps | perps/clawd-agents-perps/src, perps/clawd-agents-perps/dist, .env.example, README.md, package.json, package-lock.json, and tsconfig.json | perps/clawd-agents-perps/node_modules, live .env files, wallet/key material |
| Gateway + ADK | gateway, adk, agents, sdk, tui, and registry docs/templates | live RPC URLs, API keys, bot tokens, UCAN credentials, generated dependency folders |
Repository Galaxy
| Zone | Paths | Publish posture |
| --- | --- | --- |
| Agent interface | .claude, .github, adk, agent-kit, agents, mcp, skills | Track source and examples only; local Claude settings stay ignored |
| Gateway + governance | gateway, x402, attestation, formal_verification, docs, examples | Track public code, proofs, manifests, and templates |
| Solana execution | amm, perps, programs, pump-fun, pinocchio, llm_oracle, vulcan-cli-master | Track source and examples; never commit keys or live .env files |
| Apps + operators | apps, tui, chrome-extension, operator, clawd-cloud-os, tailclawd | Track app source; tui/.env, tui/dist, and tui/node_modules stay local-only |
| Automation + memory | automaton-main, leviathan, MemeBRain, moltbook-agent, ooda, deep-clawd, data | Track source/config templates; private data dumps stay local-only |
| SDK + packages | packages, sdk, clawdcli, clawdrouter, openShell, email-worker, scripts | Track reusable package source and automation |
| Local-only vaults | .local-secrets, openclawd-framework, UltraThink-SKill, node_modules | Ignored. Do not push to GitHub, gitlawb, npm, or deployment artifacts |
| Root docs/config | .env.example, .gitattributes, .gitignore, .gitmodules, .nvmrc, biome.json, bun.lock, bunfig.toml, CONTRIBUTING.md, HACKATHON.md, install.sh, leviathan.sh, LICENSE, MIGRATE.md, netlify.toml, package-lock.json, package.json, README.md, SECURITY.md, skills-lock.json, SOUL.md, STARTHERE, STARTHERE.md, STRATEGY.md, TRADE.md, tsconfig*.json | Track public project metadata and non-secret examples |
Viral surface area is deliberate: animated docs, explicit agent maps, private gateway manifests, verifiable Solana receipts, and x402 routes are all visible. Private material is not.
🦞👑 /Perps — The Execution Heart
OI alone is not the signal. The Clawd Perps Core OI Signal is:
OI delta + mark price delta + funding pressure + long/short skew + orderbook liquidity + spread + mark/index basis + account health = Clawd Core OI Signal → regime → score → gate → route
| Regime | Condition |
| --- | --- |
| LONG_CONTINUATION | price ↑ + OI ↑ + funding sane + spread tight |
| SHORT_CONTINUATION | price ↓ + OI ↑ + funding sane + spread tight |
| SHORTS_CLOSING | price ↑ + OI ↓ — do not chase |
| LONGS_CLOSING | price ↓ + OI ↓ — wait |
| CROWDED_LONG_RISK | OI ↑ hard + funding very positive + skew long |
| CROWDED_SHORT_RISK | OI ↑ hard + funding very negative + skew short |
| DATA_INVALID | any execution gate failed |
# OI signal — mock (no RPC needed) clawd-agents-perps signal oi SOL-PERP --mock # Live + watch loop clawd-agents-perps signal watch SOL-PERP --interval 5s --mode paper # Risk gate before entering clawd-agents-perps signal risk-gate SOL-PERP --notional 500 --side long # Strategy runners (paper-first, ledgered, gated) clawd-perps perps twap SOL --side buy --notional-usdc 500 --slices 5 --detached clawd-perps perps grid SOL --center-on-mark --width-pct 2.5 --levels-per-side 5 --tokens-per-level 0.5 clawd-perps perps finalize <run-id> --cancel-orders --close-position --wait --yes
The OI signal is the Observe layer of the perps OODA loop:
Phoenix reads the tape → RPC verifies the chain → Clawd scores the crowd → Imperial routes only after the shell says risk is clean
→ <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">perps/clawd-agents-perps/README.md</code> — full OI signal reference
→ <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">Perps/README.md</code> — full execution stack (Imperial · Phoenix · Vulcan)
⚡ MCP Server v3 — Federated Trust Machine
The MCP server is the central orchestration and receipt plane. Every tool call becomes a signed event: who called it, what model requested it, what permission allowed it, what data was touched, what settlement occurred.
curl -fsSL https://raw.githubusercontent.com/x402agent/solana-clawd/main/MCP/install.sh | bash solana-clawd-mcp # STDIO for Claude Desktop / Cursor / VS Code PORT=3001 solana-clawd-mcp-http # Streamable HTTP + SSE
{ "mcpServers": { "solana-clawd": { "command": "~/.local/bin/solana-clawd-mcp" } } }
MCP Server
┌──────────┐ ┌────────────┐ ┌──────────────┐
│ Plugin │ │ Federation │ │ Agent Task │
│ Registry │ │ Bridge │ │ Router │
└──────────┘ └────────────┘ └──────────────┘
↓ ↓ ↓
┌─────────────────────────────────────┐
│ Orchestrator + SessionMeter │
│ + PTokenStreamFacilitator │
└─────────────────────────────────────┘
| Category | Count | Description |
| --- | --- | --- |
| solana | 11 | Public Solana market data |
| helius | 8 | Helius RPC / DAS / Webhooks |
| x402 | 9 | Payment protocol + p-token metered billing |
| leviathan | 9 | OODA loop + autonomous agent control |
| market | 5 | Composite intelligence (premium) |
| pump | 8 | Pump.fun bonding curve |
| memory | 4 | Persistent agent memory + autoDream |
| agents | 6 | Agent fleet + skill management |
| deep-clawd | 6 | DeepSeek trading agent tools |
| docs | 3 | Documentation system |
| federation | N | Federated MCP tools from external servers |
| orchestrator | 6+ | Orchestrator management, integration status |
p-token stream settlement: 98.3% CU reduction vs SPL Token (6,200 → 105 CU). Atomic, batched, and streamed modes. $0.0001/token micropayments — cheap enough for high-frequency agent receipts.
🤖 Crustacean Automation (clawd-automaton)
Sovereign agent runtime — OODA loops, identity provisioning, sandbox lifecycle, replication, SQLite-backed state. All locally, all yours.
npm install -g clawd-automaton clawd-automaton --run # OODA loop clawd-automaton --goblin # Devnet paper Goblin mode
| Tier | USDC | Pulse | Posture |
| --- | --- | --- | --- |
| deep | >= $5.00 | 60s | full capability |
| shallow | >= $1.00 | 5m | economical hunting |
| shoreline | >= $0.10 | 15m | conserve resources |
| beached | $0 | — | stop before harm |
🖥️ TUI — Bloomberg-Style Sovereign Terminal
██████╗██╗ █████╗ ██╗ ██╗██████╗ ██╔════╝██║ ██╔══██╗██║ ██║██╔══██╗ ██║ ██║ ███████║██║ █╗ ██║██║ ██║ ╚██████╗███████╗██║ ██║╚███╔███╔╝██████╔╝
npm install -g @openclawdsolana/clawd-tui clawd-tui # or: hermes # Mint a real Metaplex agent clawd-agent mint --network devnet --keypair ~/.config/solana/id.json \ --name "My AI Agent" --uri https://example.com/agent.json \ --service MCP=https://example.com/mcp --yes # Gasless mint (no local SOL needed) clawd-agent mint-free --network devnet --owner <YOUR_SOLANA_PUBKEY> \ --name "My AI Agent" --uri https://example.com/agent.json
🧠 MemeBRain — Persistent Agent Memory
remember → vault note → SQLite bank → recall → action
Local-first memory: SQLite bank, markdown research vault, OODA journal ingestion, fast recall.
clawd-brain init clawd-brain remember "Jupiter Perps Risk" "Track liquidity, funding, oracle failure modes." --kind perp clawd-brain recall "Jupiter perp risk" npm run brain:ingest-ooda
📦 On-Chain Programs
| Program | Description |
| --- | --- |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">agent-minter</code> | On-chain agent minting with Metaplex integration |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">clawd-stake</code> | $CLAWD staking with reward distribution |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">llm_oracle</code> | On-chain LLM callback bridge |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">mpl-corenft-staking</code> | MPL Core NFT staking with yield mechanics |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">p-token-launchpad</code> | SIMD-0266 p-token launch and stream settlement |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">solana-ai-inference</code> | On-chain AI inference routing primitives |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">solana-gpt-oracle</code> | GPT oracle bridge for on-chain AI responses |
| <code class="bg-slate-800 text-amber-300 px-1 rounded text-xs font-mono">token-launcher</code> | Token launch automation with metadata + LP |
📦 Packages
npm i -g @openclawdsolana/clawd # Main operator CLI npm i -g @openclawdsolana/clawd-tui # TUI + Metaplex Agent Registry npm i -g @openclawdsolana/clawd-perps # Phoenix perps + OI signal + Vulcan npm i -g clawd-automaton # OODA runtime + dashboard npm i -g solana-clawd # Root CLI npm i @openclawdsolana/clawd-wallet # Wallet SDK npm i @openclawdsolana/clawd-sdk # On-chain SDK, curves, vaults
⚡ x402 — Agent-Native Payments
HTTP 402 Payment Required as agent-native settlement. Runtime loops that earn, pay, and keep operating. p-token micropayments cheap enough for per-action receipts.
curl https://x402.wtf/api/agents | jq . curl https://x402.wtf/registry | jq .
🤝 Backrooms
Three agents. One room. No exit. Analyst ↔ Satirist ↔ Clawd
curl -fsSL https://backrooms.x402.wtf/enter.sh | bash enter --loop 5
🗺️ Repo Map
solana-clawd/ ├── packages/clawd/src/aud-loop.ts ← AUD Loop — Algorithmic Utility Delta ├── mcp/ MCP v3 — receipt middleware + federation ├── sdk/ runtime source, goals, knowledge ├── automaton-main/ clawd-automaton OODA runtime + dashboard ├── attestation/ Solana Attestation Service + clients ├── operator/ OpenClawd Operator loop + ACP ├── agents/ 134-agent catalog + 115 skills ├── leviathan/ sovereign runtime ├── x402/ payment rails ├── MemeBRain/ SQLite + markdown memory vault ├── tui/ Bloomberg TUI + Metaplex Agent CLI ├── perps/clawd-agents-perps/ OI Signal + Imperial + Phoenix + TWAMM ├── Perps/ Phoenix · Vulcan · market-maker suite ├── programs/ Anchor programs workspace │ ├── agent-minter/ │ ├── clawd-stake/ │ ├── llm_oracle/ │ ├── p-token-launchpad/ │ ├── solana-ai-inference/ │ └── token-launcher/ ├── packages/ │ ├── clawd/ main operator CLI + AUD loop │ ├── clawd-perps/ Phoenix perps + OI signal │ ├── clawd-protocol/ vaults, staking, adaptive curves │ ├── clawd-sdk/ on-chain SDK │ ├── clawd-wallet/ wallet SDK + safeguards │ └── agentwallet/ encrypted keypair vault └── openclawd/ x402, gateway, E2B subtree
📖 Reading Order
Fast Start
git clone https://github.com/x402agent/solana-clawd.git cd solana-clawd npm install && npm run check npm run hermes npm run leviathan:spawn npm run automation:ci npm run programs:map npm run brain:init
╔═══════════════════════════════════════════════════════════════════════════╗ ║ CLAWD — The Verifiable Agentic Harness ║ ╠═══════════════════════════════════════════════════════════════════════════╣ ║ Harness clawd · leviathan · clawd-automaton · OODA ║ ║ Execution Phoenix · Vulcan · Imperial · OI Signal · AUD Loop ║ ║ Receipts Proof of Agentic Action · SAS attestation · memory journal ║ ║ Programs agent-minter · clawd-stake · llm_oracle · p-token-launchpad ║ ║ Memory MemeBRain · SQLite vault · OODA ingestion ║ ║ Payments x402 · HTTP 402 · Solana USDC · p-token stream settlement ║ ║ Agents 134 agents · 115 skills · gasless MPL Core minting ║ ╚═══════════════════════════════════════════════════════════════════════════╝
License
MIT.
backrooms.x402.wtf · x402.wtf · solanaclawd.com · the shell molts. the laws do not.