Ctrlk
For the complete documentation index, see llms.txt. This page is also available as Markdown.

Introduction

The runtime layer for AI.

Your AI remembers, keeps working, and survives every session.

Most AI assistants disappear when the conversation ends. Noelclaw gives them persistent state - memory that accumulates, agents that keep running, vaults that version knowledge, and workflows that continue after you close the chat.

Works in Cursor, Windsurf, Claude Desktop, ChatGPT, Zed, Hermes, Bankr, Aeon, and anywhere MCP runs.

The three pillars

Memory - what your AI remembers

Semantic, versioned, deduplicated. Vault gives you git-style versioning and knowledge-graph links. Memory gives you semantic search with 90-day time-decay and same-session deduplication. Together they form a two-tier persistent state your AI can read from and write to.

Agents - what runs in the background

Named, persistent agents that survive across sessions. Spawn one with a goal, recall it weeks later, audit every state change via the ledger. Each agent can hold its own Base wallet address for on-chain identity.

Workflows - what executes on a schedule

Packets, automations, monitors, and deep research. Anything that runs after you close the chat - daily research that lands in your vault, DCA orders that fire weekly, multi-agent research swarms that complete async.

Quick Install

Requirement: Node.js >= 18 - check with node --version, download from nodejs.org if needed.

One-command setup (auto-detects all MCP clients)

Detects Claude Desktop, Cursor, Windsurf, VS Code, Zed, and configures each automatically. Then restart your client.

Claude Code

Claude Desktop

Edit your config file:

  • Mac: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop after saving.

Hermes

No build step. No config required. Runs on first use.

121 Tools

Noel Shell: The webapp chat supports native tool calling. Shell tools let the chat spawn agents, save to vault, search memory, create automations, estimate + execute swaps, list agents, and check wallet balances — all from natural conversation. See Noel Shell.

Grouped by pillar — counts match the live tool registry exactly. Full catalog: All 121 Tools.

Pillar
Category
Tools
What it does

Memory

Vault

15

Versioned notes: save, read, search, history, diff, export, credentials, tags, links, knowledge graph

Memory

Semantic Memory

10

Semantic search, cross-session recall, extract, consolidate, insight, publish/delete with confirm

Memory

Chronicle

4

Append-only audit trail — add, list, search, activity stats

Agents

Agents

12

Persistent named agents: spawn, recall, update, identity, ledger, schedule/pause/resume, runs, hire, list

Agents

Playbooks

3

Browse and run curated playbooks with an audit ledger

Workflows

Automations

6

DCA, price alerts, conditional buy/sell, dry-run, pause/delete, execution history

Workflows

Monitors

3

Scheduled research monitors: schedule, list, cancel

Workflows

Packets (Flows)

4

Create, run, list, and share reusable workflow packets

Workflows

Research

3

deep_research (keyless evidence packs), compare two reports, walk a research chain

Execution

Base DeFi

12

base_mcp_*: balances with impostor detection, send, 0x Permit2 swaps, Morpho/Moonwell yields, basenames — see Base

Execution

Robinhood Chain

14

rh_*: 22 tokenized stocks + any RH-chain crypto, V2/V3/V4 best-fill routing, risk screens, DCA/TP/SL orders, stock bridge — see Robinhood Chain

Execution

Stocks (SEC)

3

Fundamentals (XBRL), insider trades (Form 4), material events (8-K) — see Stocks & SEC Data

Execution

Market Data

6

Live prices, token data, market overview, OHLC history, market thesis

Execution

Token Scanner

3

Score, safety-check, and compare tokens on Base

Execution

Web Research

2

Live web search + scrape any URL (proxied, keyless)

Execution

GitHub

8

Repos, PRs, issues, files, commits, code search (read-only)

Execution

Code Audit

1

Static Solidity security scan with review rubric

Execution

Insight

3

ask_noel, trade planning, DeFi yields

Execution

MiroShark

3

Multi-agent market simulation

Execution

Wallet

3

Wallet address, live balance with USD pricing, EIP-191 message signing

Runtime

OS

3

noel_status dashboard, noel_diagnostics health check, Noel Shell chat bridge

What's New

v3.43.1 - x402 API Market, Robinhood Chain Trading, 121 Tools

  • x402 API Market — sell any HTTPS API per call, buy with USDC on Base via the x402 protocol (HTTP 402 challenge → pay → retry with PAYMENT-SIGNATURE). No account or API key needed for buyers. Payments are replay-protected at both the request-id and tx-hash level. See x402 API Market.

  • Robinhood Chain rail (rh_*) — 22 tokenized stocks (added NFLX, SPY, QQQ, GME) plus trending chain tokens, arbitrary crypto by ticker or contract address, V2/V3/V4 best-fill routing, receipt-backed swap confirmations, safety scans (rh_analyze, rh_safety_check), DCA/TP-SL order engine.

  • Investor triad from SEC primary sources, all keyless — stock_fundamentals (XBRL financials), stock_insider (Form 4), stock_events (8-K decoded), plus rh_stock_bridge (tokenized vs real price + pool depth).

  • Client-first refactor — tools return evidence and structure; your model does the reasoning. deep_research works with zero API keys (mode: "sources").

  • Cleaner metadata — tool count corrected to 121 everywhere, emoji removed from all tool and package descriptions.

v3.32.7 - Local Memory, OpenAI BYOK, Critical Install Fix

  • Local memory — run memory tools on a free, self-hosted supermemory server on your own machine. Zero cost, private, no Noelclaw account or Convex proxy needed once enabled. Run noelclaw setup to switch. (Beta - code-reviewed and unit-verified, live-server testing pending.)

  • OpenAI BYOK — OpenAI joins Bankr/Anthropic/Grok as a direct LLM provider. OPENAI_BASE_URL lets you route to any self-hosted OpenAI-compatible gateway instead (LiteLLM, vLLM, Ollama, OpenRouter, your own VPS).

  • noelclaw setup — new guided CLI wizard for picking an LLM provider and/or enabling local memory in one flow.

  • Critical fix — noelclaw install was writing a broken MCP server entry (npx -y @noelclaw/mcp@latest, ambiguous bin resolution) into every detected client config, so every fresh auto-install was non-functional. Fixed to the unambiguous, version-pinned form.

v3.32.5 - Chronicle Search, execute_swap, Live Wallet Pricing, Diagnostics

  • execute_swap — execute token swaps on Base mainnet from Noel Shell. Enforces estimate → confirm → execute flow. Routes via 0x Permit2. Returns tx hash + Basescan link. Hard-blocked without confirmed=true.

  • chronicle_search — keyword search across runtime events by title and detail. Find when your agent last researched any topic without scrolling the full log.

  • chronicle_stats — runtime activity analytics: event breakdown by type, daily heatmap, busiest days, avg events/day over a configurable window (default 30 days, max 90).

  • get_wallet_balance — live ETH + USDC balance from Base mainnet with real-time USD pricing from CoinGecko. No API key required.

  • wallet_sign_message — EIP-191 personal_sign to prove wallet ownership off-chain without sending a transaction.

  • noel_diagnostics — pre-flight health check: pings Convex, Firecrawl, Supermemory; lists configured API keys; warns on missing LLM key or Firecrawl key with actionable hints.

  • Base Builder Code — all x402 payment flows now include bc_7diuqbqo as builderCode. Transactions are attributed to Noelclaw in the Base Dashboard.

v3.31.0 - Noel Shell, Multi-Provider Chat, Security Hardening

  • Noel Shell - native tool calling from the webapp chat. Shell tools: spawn_agent, save_to_vault, search_memory, create_automation, estimate_swap, list_agents, get_wallet_balance. The chat can now act, not just answer. See Noel Shell docs.

  • 7 Agents - Noel (crypto), CoinGecko (crypto data), Sage (analysis), Forge (developer), Quill (creative), Spectre (trading), Atlas (general). Each agent has its own persona and tool access.

  • Multi-provider chat - provider cascade: Bankr → OpenAI → Anthropic → Groq → OpenRouter → Custom → Local fallback. No single provider dependency.

  • ConnectMcpModal - onboarding flow for connecting the MCP server to your IDE directly from the webapp.

  • Security hardening - 8 security boundaries enforced + vulnerability fixes: getDecryptedPKByUserId → internalAction, createWallet → internalAction, getPrivateKey returns address only (never raw key). Auth is Privy + API key only (OTP removed Jul 2026).

  • Theme refresh - Claude-style warm palette, Inter font, neural-network knowledge-graph visual.

v3.23.1 - Polish

  • Memory dedup - memory_add deduplicates identical content via SHA-256 hash with in-process LRU cache + recent-memories lookup. force: true overrides.

  • Agent race-safe - concurrent agent_update calls on the same agent serialize through a per-name async mutex. No more silent write loss.

  • Automation dry-run + error categories - run_automation dryRun: true simulates without broadcasting. Failed runs show category badges (INSUFFICIENT_BALANCE, QUOTE_FAILED, TX_REVERTED, etc.) with one-line fix suggestions.

  • MCP Resources pagination - vault entries past the 50th are now visible to clients via cursor pagination. MIME types derive from contentType (markdown/json/code/text).

v3.22.0 - UX

  • HTTP cache + 429 backoff - every external API call (CoinGecko, DexScreener, GeckoTerminal) flows through a 45s LRU + exponential backoff that honors Retry-After. Agent loops can't trip rate limits anymore.

  • MEV-protect RPC opt-in - NOELCLAW_BROADCAST_RPC=<private-relay-url> routes signed transactions through a private relay. Reads stay on the fast Base RPC.

  • Tool count drift fixed - banner, login, and noelclaw doctor all derive counts from the actual registered tools. No more 104/110/36/100+ mismatch.

  • Humanizer model pin - NOELCLAW_HUMANIZER_MODEL lets you lock the model used by humanize_text and write_content for consistent voice.

v3.21.0 - Safety

  • Slippage + price-impact guards - swap_tokens and base_mcp_swap refuse execution when price impact exceeds maxPriceImpactPct (default 3%). Default slippage cap 1%. Configurable per call.

  • Silent supermemory sync fixed - retries 3× exponential, logs to chronicle on permanent failure. No more ghost memories that exist in one tier but not the other.

  • SWARM_TOOLS zombie removed - dead dispatch handler cleaned up. Multi-agent research is built into deep_research now (depth: "standard" | "deep").

v3.20.0 - Grounding & streaming

  • audit_contract grounded - 13-pattern static Solidity scan runs before the LLM, mandatory disclaimer appended, no more "secure"/"safe" claims.

  • github_search_code helpful error - clear setup instructions and a link to https://github.com/settings/tokens when GITHUB_TOKEN is missing.

  • memory_search time-decay - 90-day half-life weighting + promotion hints when 4+ memories cluster on a topic.

  • Pyth oracle in fetchVerifiedPrice - cross-source price verification against CoinGecko + DexScreener + Pyth with disagreement detection.

  • deep_research streaming progress - MCP notifications/progress for long-running multi-agent research.

  • noelclaw doctor - 5-second CLI health check with ✓/⚠/✗ status and inline fix hints.

Docs

NextQuickstart

Last updated

Was this helpful?