Scout
Detects opportunities — new listings, launches, emerging momentum.
Detect at the source
Clawlas is a personal AI-assisted system for detecting, evaluating, simulating and trading newly launched tokens and new listings — seconds-old contracts, first liquidity, bonding curves, DEX migrations and CEX listings. A team of specialized agents proposes; the human governs. Every decision is explainable, measurable and remembered.
- Early-token intelligenceSpecialized agents — early discovery, launch analysis, quant, security, wallet-intelligence, on-chain, social, debate, risk, memory — detect tokens from their first seconds and turn intuition into measurable evidence, source-first.
- Verify the exit, not just the entryA mandatory sellability layer proves a token can likely be sold — not just bought. Honeypots, sell taxes, blacklists and vanishing liquidity are hard blocking rules no momentum score can override.
- Shadow mode first, no money at riskBefore a cent is risked the platform only observes: it records what it would buy, at what price and why, then compares against reality from 5 seconds to 30 days. Profitable patterns are proven before capital follows.
Sign in
Sign in to your account
The research team
A team of specialized agents
The AI is not one model making a call — each agent owns a narrow job and feeds evidence into the scoring engine, with a Memory agent threading knowledge across decisions.
Research
Builds complete investment reports for a candidate.
Quant
Creates features, indicators and statistical models.
Security
Analyzes contracts and rug-pull risk.
On-chain
Studies wallets, liquidity and blockchain activity.
Social
Measures narratives and community momentum.
Debate
Attempts to invalidate a proposed trade.
Risk
Protects capital.
Memory
Maintains long-term knowledge across decisions.
Governance & safety
The platform proposes, the human governs.
The multi-agent system explores, scores and proposes continuously — but every move that reshapes the lab or reaches the market belongs to the human. The boundary is hard-coded in the tool registry, never in a prompt.
Research runs autonomously
Detection, scoring and shadow-mode observations run continuously, no approval needed — none of it risks capital.
Capital & rules need a human
Strategy, deployment, capital allocation, execution permissions and scoring changes are manual, human-validated decisions.
No arbitrary orders
There is no arbitrary order-placement tool for the LLM to call — agents act only within hard-coded tool bounds.
No-real-money key in research
Through shadow mode and simulation a no-real-money key keeps real funds physically unreachable — even a compromised agent reaches nothing.
Hard guardrails
- LLM output is untrusted: it never runs shell, queries secrets, submits orders or changes risk limits.
- Tool calls pass through allowlists, schema validation, authorization and deterministic handlers.
- Exchange keys are code-only — never in Convex, prompts or logs; trading-only, withdrawals disabled, IP-restricted.
- An append-only audit log records every strategy change, approval, score change, order and intervention.
The phases
From foundation to controlled production
Eleven phases take the lab from infrastructure to live execution — trading only begins once detection, security, features and agents exist, and even then through shadow → paper → minimal capital.
Build progress2 of 11 phases shipped · 18%
In progress: Phase 3 · Integrations & observability foundation
Phase 1 · Foundation
Monorepo, web app, Convex, FastAPI, VPS + Traefik, CI/CD, logs & monitoring, tables
Phase 2 · Application shell & console
Authenticated app shell: collapsible grouped sidebar, profile + logout, account/settings, dashboard & charts, dark mode
Phase 3 · Integrations & observability foundation
External connectors, secrets, provider budgets, AI/API usage logs, model registry, cost dashboards
Phase 4 · Early token detection
Chain / launchpad / pool collectors, first-liquidity & first-trade, lifecycle engine, tables & Parquet
Phase 5 · Security & sellability
Deterministic scanners, sell simulation, hard blockers — before any AI synthesis
Phase 6 · Quantitative core & features
Deterministic versioned formula engine, feature registry, Signal/Risk scores
Phase 7 · Agent research
LangGraph workflows, launch/wallet/social agents, memory, debate & risk, AI score
Phase 8 · Model routing & cost governance
Deterministic task detection, per-task benchmarks, validated hybrid routing, cost caps
Phase 9 · Shadow exploration
Hypothetical decisions on live data, early snapshots, multi-window outcomes, weekly reports
Phase 10 · Paper trading
Realistic event-driven execution simulation + strategy competition
Phase 11 · Controlled production
Isolated execution, minimal capital, kill switch, reconciliation, human governance
The stack
Five layers, one source of truth
Every layer is isolated behind a public interface and validated at its edge — pick the slice you’re working in without learning the whole tree.
Frontend
TanStack Start
Type-safe routing and server functions, rendered with React and styled with Tailwind.
- TanStack Start
- TanStack Query
- React 19
- Tailwind v4
- shadcn/ui
Realtime backend
Convex
Queries, mutations and actions on a reactive database — authorize first, validate every arg.
- Convex
- Reactive queries
- Server actions
Auth
Better Auth
Email + password sessions, isolated behind a single useAuth / requireUser interface.
- @convex-dev/better-auth
- requireUser
- AuthBoundary
Shared contract
Zod schemas
One schema, inferred on both sides — the client and Convex never disagree on a shape.
- Zod
- Inferred types
- react-hook-form
Tooling
Turborepo
pnpm workspaces, ESLint boundaries, Vitest and Prettier — the guard-rails ship wired up.
- pnpm
- Turborepo
- ESLint boundaries
- Vitest
- Husky
One command
From clone to running in one step
pnpm bootstrap provisions everything — install, link Convex, push secrets and write the env file — then pnpm dev brings the whole stack up.
$pnpm bootstrap
✓dependencies installed
✓Convex deployment linked
✓secrets set · apps/web/.env.local written
$pnpm dev
→http://localhost:3000
Go deeper
Explore the documentation
Every layer, agent and decision is documented — from the early-token pipeline to the cost model. Start anywhere.
System architecture
How the vertical slices fit together, type-safe from the client to Convex.
Early tokens
The opportunity categories — launches, bonding curves, migrations, CEX listings.
Multi-agent system
The specialized agents that turn opportunities into measurable evidence.
Sellability layer
Proving a token can be sold — honeypots, sell taxes and vanishing liquidity.
Risk engine
Hard blockers and exposure limits that protect capital.
Human governance
What the platform proposes on its own, and what only the human governs.
Cost model
LLM pricing, allocated vs cash cost, and the budget phases.
Implementation phases
The eight build phases from foundation to controlled production.