EchoClaw CLI

One CLI surface for building and operating on the 0G ecosystem from terminal or AI agents. The core principle is simple: if a framework can execute CLI commands, it can use EchoClaw.

What This Package Is Today

As of March 7, 2026, EchoClaw is in a practical hybrid state:

• Provider-agnostic core exists across wallet, transfers, Jaine, ChainScan, EchoBook, Slop modules, MarketMaker runtime, and update runtime.
• Multi-provider skill linking exists (openclaw, claude, claude-code, codex, other).
• Claude Code now has a dedicated local backend path (guided flow, proxy, config inject/remove/restore).
• OpenClaw-specific integration is still the deepest agent/gateway path (guided launcher flow, hooks/gateway wiring, compute config patching, restart guidance).

Candor on boundaries:

• If you are not using OpenClaw, core CLI commands still work.
• OpenClaw hook/gateway flows are not provider-agnostic and require OpenClaw runtime to deliver notifications.

Install

npm install -g @echoclaw/echo-0g

Requires Node.js >= 20.

Start Here

For humans, the supported entrypoint is now:

echo0g echo

echo0g echo is the guided launcher for:

• connecting OpenClaw, Claude Code, Codex, or another agent runtime
• funding your shared 0G compute wallet / ledger / provider balance
• exploring safe read-only parts of the ecosystem
• checking status, diagnostics, and support data
• handling Claude config remove / restore / proxy lifecycle from one place

The first screen is task-first:

1. Connect my AI
2. Fund my AI in 0G
3. Bridge / Cross-Chain
4. Explore Echo
5. Wallet & Keys
6. Manage / Fix
7. Advanced

Runtime Notes

OpenClaw

Use echo0g echo, then choose Connect my AI -> OpenClaw.

This path gives you:

• OpenClaw skill linking and environment wiring
• hook routing configuration through the existing setup openclaw-hooks flow
• compute funding + monitoring guidance
• restart guidance when env/config changes need a gateway reload

Important boundary:

• monitor notifications depend on OpenClaw-compatible hook routing
• the launcher only offers monitor notifications when the selected runtime is OpenClaw

Claude Code

Use echo0g echo, then choose Connect my AI -> Claude Code.

This path gives you a local Claude Code backend on top of 0G Compute:

• guided provider/model selection
• funding, ACK, and API key creation
• local Claude settings injection
• local translation proxy on 127.0.0.1
• remove / restore flow for Claude config

Default Claude integration target:

• project-local .claude/settings.local.json

Advanced follow-up commands stay available under the new root:

echo0g echo claude proxy status
echo0g echo claude config show
echo0g echo claude config inject --scope project-local|user|project-shared
echo0g echo claude config remove
echo0g echo claude config restore

Important integration details:

• EchoClaw merges only its managed Claude keys into the target settings file; it does not replace the whole file.
• Injected aliases pin Claude sonnet, opus, haiku, and subagents to the selected 0G-<provider-model> label.
• remove removes only untouched EchoClaw-managed keys; restore replays the full pre-injection snapshot.
• Context compaction remains Claude Code behavior; EchoClaw provides the local gateway path and token-count estimate.

Codex / Other frameworks

Use echo0g echo, then choose Connect my AI -> Codex or Other.

Human-first path:

• the launcher links the skill when auto-linking is supported
• if auto-link is not possible, it shows the exact manual source path and next steps

Low-level agent surface still exists:

echo0g skill install --provider codex|other --scope user|project
echo0g skill path --json

AI Agent Platforms (MVP)

| Platform | Install targets (--scope user / --scope project) | Auto-link behavior | Current support | |---|---|---|---| | OpenClaw | ~/.openclaw/skills/echo0g (+ workspace mirror), project: <cwd>/skills/echo0g | Linked automatically | Full integration: guided launcher path, hooks, monitoring, automation | | Claude Code | ~/.claude/skills/echo0g, project: <cwd>/.claude/skills/echo0g | Linked automatically | Guided local backend: alias/subagent pinning, proxy, config inject/remove/restore | | Codex | ~/.agents/skills/echo0g, project: <cwd>/.agents/skills/echo0g | Linked automatically | Skill linking + manual CLI commands | | Other | Source path only (returned by CLI) | Always manual_required (no auto-link) | Manual fallback path output |

Human-first vs agent-first

EchoClaw now deliberately separates:

• human-first: echo0g echo
• agent-first: echo0g echo ... --json task flows plus low-level module commands That means:
• people should start with echo0g echo
• agents should prefer the task-first planner/executor surface when the user asks for setup, repair, funding, or verification:

echo0g echo connect --plan --json
echo0g echo connect --apply --runtime claude-code --json
echo0g echo fund --plan --json
echo0g echo fund --apply --provider <addr> --amount 1 --json
echo0g echo verify --runtime openclaw --json
echo0g echo status --json
echo0g echo doctor --json
echo0g echo support-report --json

• skills and automations can still call explicit low-level commands like wallet, 0g-compute, jaine, chainscan, slop, echobook, 0g-storage, and skill install
• the old public runtime roots onboard, openclaw, claude, and 0g-compute wizard/openclaw are no longer the supported human entrypoints

0G Compute: How It Fits

0G Compute itself is a decentralized inference network (official overview): https://docs.0g.ai/developer-hub/building-on-0g/compute-network/overview

For humans, start with echo0g echo and choose Fund my AI in 0G.

That flow covers:

• live provider/model discovery
• pricing-aware recommendations (min locked 0G)
• ledger deposit and provider funding
• signer ACK + API key creation
• live balance refresh without stale broker cache
• optional monitor start/stop when the runtime is OpenClaw

The low-level compute surface remains available for agents and advanced users:

echo0g 0g-compute setup --json
echo0g 0g-compute providers --detailed --with-balances --fresh --json
echo0g 0g-compute provider <address> info --fresh --json
echo0g 0g-compute ledger status --json
echo0g 0g-compute monitor start --daemon

Important runtime boundary:

• monitor notifications depend on OPENCLAW_HOOKS_* routing and an OpenClaw-compatible gateway
• Claude Code uses a local Anthropic-to-OpenAI proxy because the live 0G broker currently exposes OpenAI /chat/completions, not Anthropic /v1/messages, on the deployed path
• Codex/other runtimes still use skill linking plus direct CLI orchestration

Core Module Surface

| Module | What it does | |---|---| | Wallet | Create/import wallet, inspect native and Khalani-backed balances | | Transfers | Native 0G transfer with prepare -> confirm intent flow | | Jaine DEX | Pools, swaps, allowances, LP operations | | Jaine Subgraph | Read-only DEX intelligence and analytics | | 0G Compute | Provider/account setup, funding, API keys, readiness checks | | MarketMaker | Trigger-based trading daemon | | Update | One-shot package auto-update on CLI use | | ChainScan | On-chain data queries | | EchoBook | Social feature command set | | Slop Bonding | On-chain token lifecycle commands on Slop bonding curve | | Slop Stream | Real-time token stream ingestion | | Slop App | slop.money app-layer API commands | | 0G Storage | Durable agent storage: file upload/download, virtual drive, notes, snapshots |

Most Used Commands

# Human-first launcher
echo0g echo
echo0g echo connect --plan --json
echo0g echo connect --apply --runtime openclaw --json
echo0g echo fund --plan --json
echo0g echo fund --apply --provider <addr> --amount 1 --json
echo0g echo verify --runtime claude-code --json
echo0g echo status --json
echo0g echo doctor --json
echo0g echo support-report --json

# Claude Code local backend
echo0g echo claude
echo0g echo claude proxy status
echo0g echo claude config show

# Skill linking (multi-provider)
echo0g skill install --provider openclaw|claude|codex|other

# Wallet + transfers
echo0g wallet ensure --json
echo0g wallet balance --json
echo0g wallet balances --json
echo0g wallet address --chain solana --json
echo0g send prepare --to <address> --amount <0G> --json
echo0g send confirm <intentId> --yes --json

# Khalani / cross-chain
echo0g khalani chains --json
echo0g khalani tokens top --json
echo0g khalani quote --from-chain eth --to-chain base --from-token 0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48 --to-token 0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913 --amount 1000000 --json
echo0g khalani quote --from-chain eth --to-chain base --from-token 0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48 --to-token 0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913 --amount 1000000 --stream --json
echo0g khalani bridge --from-chain eth --to-chain base --from-token 0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48 --to-token 0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913 --amount 1000000 --dry-run --json
echo0g khalani orders --json

# Jaine
echo0g jaine pools scan-core --json
echo0g jaine swap sell w0G USDC --amount-in 1 --dry-run --json
echo0g jaine lp list --json

# Automation
echo0g marketmaker order add --json
echo0g marketmaker start --daemon
echo0g marketmaker status --json

# Compute monitor
echo0g 0g-compute monitor start --daemon

# Updates
echo0g update enable
echo0g update status --json
echo0g update check --json

Hooks, Notifications, and Daemons

OpenClaw-only boundary

setup openclaw-hooks and setup test-hooks are designed for OpenClaw gateway architecture.

echo0g setup openclaw-hooks --base-url <url> --token <secret> --channel <ch> --to <recipient>
echo0g setup test-hooks --json
echo0g setup test-hooks --probe-live --json

What this means for non-OpenClaw setups:

• Long-lived daemon webhook notifications depend on OPENCLAW_HOOKS_* routing and OpenClaw-compatible /hooks/agent endpoint.
• Running daemons from Claude/Codex still requires an OpenClaw gateway (or equivalent compatible endpoint) for those notifications to be delivered.

Advanced diagnostics:

• setup test-hooks --json: dry validation (token sync, gateway-token consistency checks, routing checks, payload construction).
• setup test-hooks --probe-live --json: sends real webhook probes to gateway and can create remote session turns.

Gateway Restart Required (Manual Setup)

When setup password or setup openclaw-hooks writes env/config changes, restart is required so OpenClaw host/container reloads them.

Host flow:

openclaw gateway restart
# then send /restart in chat

Container flow:

docker compose -f ~/openclaw/docker-compose.yml restart
echo0g 0g-compute monitor start --from-state --daemon
# then send /restart in chat

Auto-Update

Auto-update is enabled by default. On the first ordinary CLI command (everything except echo and update), including --json and other headless/non-TTY runs, ECHO_AUTO_UPDATE=1 is written to ~/.config/echo0g/.env (+ OpenClaw mirror) if no explicit preference exists. The update check then runs as a one-shot background install and uses npm install -g without blocking the command.

Preference commands:

• echo0g update enable — turns auto-update on for CLI use, including headless
• echo0g update disable — turns auto-update off for CLI use
• echo0g update check — read-only manual version check
• echo0g update status — shows current preference, last check state, and any legacy artifacts

Compatibility aliases:

• echo0g update start -> echo0g update enable
• echo0g update stop -> echo0g update disable

Opt-out paths:

• echo0g update disable — prevents automatic package install
• ECHO_AUTO_UPDATE=0 in .env — explicit opt-out, not overridden by default seeding
• ECHO_DISABLE_UPDATE_CHECK=1 — disables the entire update check flow (not just auto-install)

Headless mode (--json and non-TTY) now follows the same default seeding and one-shot auto-update flow. You do not need echo0g update enable just because a workflow is automated. Bootstrap-exempt roots (echo, update) still skip seeding/checking, and ordinary commands skip default seeding only when an explicit preference exists, ECHO_DISABLE_UPDATE_CHECK=1, or legacy update.stopped is migrated first.

Commands that manage update preferences (echo, update) skip auto-update seeding and checking in their preAction to avoid side effects before the user makes a choice.

Legacy note:

• update.stopped is migrated best-effort into ECHO_AUTO_UPDATE=0 during ordinary-command CLI bootstrap.
• update.pid, update.shutdown, and update-state.json are legacy cleanup artifacts only. The update daemon is no longer part of the runtime model.

Daemon Lifecycle and Resurrection

Two daemon families use PID/state/stopped marker files:

| Daemon | Files | Auto-resurrection condition | |---|---|---| | BalanceMonitor | monitor.pid, monitor-state.json, monitor.stopped | monitor-state.json exists and no monitor.stopped | | MarketMaker | bot.pid, orders.json, bot.stopped | orders.json exists with at least one armed order and no bot.stopped |

Marker semantics:

• Explicit start clears .stopped.
• Explicit stop (including SIGTERM/shutdown-file/SIGKILL stop paths) writes .stopped.
• Daemons are auto-respawned only when runtime says they should be running and marker state allows it.

Auto-update is separate:

• package updates are not a daemon family anymore,
• the updater runs one-shot on CLI use,
• preference lives in ECHO_AUTO_UPDATE / ECHO_DISABLE_UPDATE_CHECK.

Developer detail:

• Child daemon spawns use ECHO_NO_RESURRECT=1 to prevent recursive resurrection loops.
• Auto-update preference is centralized in src/update/auto-update-preference.ts — single source of truth for isAutoUpdateEnabled().

Security Model

| Feature | Current behavior | |---|---| | Password read precedence | process.env -> ~/.config/echo0g/.env -> ~/.openclaw/.env mirror -> legacy openclaw.json | | Password sanitization | Empty string and literal "undefined" are treated as missing | | Password write default | Dual-write to app .env (chmod 600) + OpenClaw .env mirror | | Password legacy path | setup password --to-config writes to openclaw.json (legacy/deprecated) | | Wallet key encryption | AES-256-GCM with scrypt KDF | | Transfer safety | Mandatory 2-step prepare -> confirm | | Headless guardrails | echo requires TTY; wallet mutations blocked by default unless ECHO_ALLOW_WALLET_MUTATION=1; interactive setup password prompt blocked unless --from-env or --password | | On-chain confirmations | Chain-writing 0g-compute actions require explicit --yes | | Secret handling | Tokens/keys are masked in output/log paths | | Writes | Atomic write strategy for config/env updates |

Troubleshooting

Common quick fixes:

• Stale PID daemon errors (monitor/marketmaker): run matching stop first to clear stale PID path, then run start.
• Hook token mismatch: echo0g setup openclaw-hooks --token <tok> --force.
• Webhooks accepted but not delivered: ensure both OPENCLAW_HOOKS_CHANNEL and OPENCLAW_HOOKS_TO are configured.

V2 / Deferred

The Khalani cross-chain surface is being added in staged form. A few parts are intentionally deferred until the execution contract is clearer or the extra complexity is justified.

• PERMIT2 live execute
  • v1 supports quote/build visibility and dry-run only.
  • We are not enabling live CLI execution until Khalani exposes a canonical execution payload or ABI contract for safe submit/broadcast without guessing.
• Solana TRANSFER deposit method
  • v1 targets the standard Solana CONTRACT_CALL path returned by Khalani.
  • Native support for SPL-style transfer deposits can be added later if real routes require it.
• Advanced custom RPC overrides per chain
  • v1 plans to use RPC metadata returned by Khalani chain definitions.
  • Per-chain manual overrides should only be added if real reliability issues appear with public RPCs.

Priority for v1: stable quote -> build -> execute -> order tracking across EVM and Solana, not maximum coverage of every deposit variant on day one.

Already implemented in v1:

• streaming quotes via echo0g khalani quote --stream --json
• NDJSON output with type: "route" events and a final type: "complete" summary
• multi-chain wallet balances with Khalani token balances plus live native balances

Current Skill Docs (Only What Exists Now)

Use only these packaged skill docs as authoritative right now:

| Skill file | Purpose | |---|---| | skills/echo0g/SKILL.md | Router skill: baseline execution rules and module selection | | skills/echo0g/references/wallet-transfers.md | Wallet lifecycle, transfer flow, config, guardrails | | skills/echo0g/references/jaine-dex.md | Jaine execution commands (swap/LP/allowance/w0G) | | skills/echo0g/references/jaine-subgraph.md | Read-only market analytics via subgraph | | skills/echo0g/references/slop-bonding.md | Bonding-curve token lifecycle and trading flow | | skills/echo0g/references/0g-compute.md | Provider setup, funding, API keys, monitor workflow | | skills/echo0g/references/marketmaker.md | Trigger-based daemon trading and operations | | skills/echo0g/references/slop-stream.md | Real-time token stream events | | skills/echo0g/references/echobook.md | EchoBook social command surface | | skills/echo0g/references/chainscan.md | Explorer-style on-chain intelligence | | skills/echo0g/references/slop-app.md | slop.money app profile/chat/agent-query layer | | skills/echo0g/references/0g-storage.md | Durable storage: file upload/download, virtual drive, notes, snapshots | | skills/echo0g/references/khalani-cross-chain.md | Multi-chain chain/token discovery, quotes, bridge execution, order tracking, and Solana wallet handling |

Notes:

• skills/echo0g/SKILL_old.md is a legacy snapshot.

For AI Agents

Use machine-readable mode whenever possible:

echo0g <command> --json

Find packaged skill location:

echo0g skill path --json

Documentation

• EchoClaw docs: https://echoclaw.ai/docs/overview/quick-start
• 0G Compute overview: https://docs.0g.ai/developer-hub/building-on-0g/compute-network/overview

Development

npm install
npm run build
npm test

License

MIT