Between January 9 and February 9, we shipped what amounts to a new platform. Not incremental improvements — a fundamental rethinking of how agents reason, how the infrastructure scales, and how much data you have at your fingertips. This is the release that turns COD3X from a capable trading tool into a system that adapts to how you trade.
Here's what changed, why it matters, and what you can do with it.
Four Ways to Think: Agent Reasoning Modes#
Not every trading decision deserves the same cognitive overhead. A simple limit order doesn't need the same reasoning depth as a multi-asset portfolio rebalance. Until now, every agent used the same approach for everything. That's over.
You now choose from four reasoning modes per agent:
Direct Mode#
When to use it: Routine operations — placing a predefined order, fetching a price, logging a journal entry. You already know what you want; the agent just needs to execute.
What it does: Skips intermediate reasoning entirely and executes the task immediately. Fastest response time, lowest token cost.
Chain-of-Thought (CoT)#
When to use it: Complex decisions that benefit from showing their work — analyzing whether to enter a position based on multiple timeframe confluence, or evaluating a trade setup against your strategy rules.
What it does: The agent reasons step-by-step, documenting each inference before arriving at a conclusion. You can review its logic and catch flawed reasoning before it acts.
Graph-of-Thought (GoT)#
When to use it: Decisions with multiple valid approaches — comparing a momentum entry vs. a mean-reversion entry on the same setup, or evaluating three different exit strategies simultaneously.
What it does: The agent explores multiple reasoning paths in parallel, scores each path, and converges on the strongest one. Think of it as the agent debating itself and picking the winner.
ReAct (Reasoning + Acting)#
When to use it: Open-ended research and multi-step operations — "scan the top 20 Hyperliquid pairs, identify the three best setups, then size and place entries." Tasks where the agent needs to observe, adjust, and act in a loop.
What it does: The agent cycles through observe → reason → act repeatedly, using each action's result to inform the next step. It adapts in real time rather than committing to a plan upfront.
Match the reasoning mode to the task. Direct for execution, CoT for analysis, GoT for comparison, ReAct for exploration. The result: better decisions, faster responses, and less wasted compute.
Agent Templates: Build Once, Deploy Many#
Templates were one of the most requested features — and for good reason. Every agent you've built from scratch represents hours of configuration: selecting tools, tuning parameters, choosing the right model, setting up goals. Doing that once is productive. Doing it repeatedly is a waste.
You can now save any agent configuration as a template and deploy copies instantly. Each template captures the full setup — reasoning mode, model assignment, tool selection, parameters, and goals. The new templates panel lets you create, browse, edit, and deploy in a few clicks.
Dynamic Tool Selection#
A subtler but equally impactful change: agents no longer load every available tool on every invocation. The system now intelligently selects only the tools relevant to the current task. This reduces token overhead (fewer tool descriptions in the prompt), improves response quality (less noise for the model to parse), and speeds up execution.
Streaming & Model Upgrades#
- Improved chat streaming — Better resume logic when connections drop, plus a new tool-call component that shows you exactly what the agent is doing in real time
- Per-bot model configuration — Each bot runs on its own AI model rather than sharing a global setting. Use a fast model for your market scanner and a powerful model for your portfolio manager.
- AI model benchmarking — New infrastructure for tracking model performance across task types, so you can make data-driven decisions about which model to assign where
Trading & Charts#
ADX: Trend Strength, Quantified#
The most requested indicator is live. The Average Directional Index measures how strong a trend is regardless of direction — a reading above 25 signals a strong trend, below 20 signals a range.
What makes our implementation different: ADX isn't just a visual overlay. It's wired into the trigger system, so your agents can act on trend strength programmatically. Set a goal that only fires when ADX crosses above 25, or one that pauses trading when ADX drops below 15 and the market goes choppy. Combine it with directional indicators for strategies that only trade strong trends in a confirmed direction.
Trading Journal#
Your best trade is useless if you can't remember why you took it. The new trading journal creates structured entries linked to positions and orders — capturing your reasoning, the market context, and the outcome in one place.
Over time, this becomes a searchable decision archive. Your agents can reference past entries to improve future decisions, and you get an audit trail that shows exactly why every position was opened, modified, or closed. No more reconstructing your logic from trade history alone.
Deeper Hyperliquid Data#
This release significantly expands what you can see and do on Hyperliquid:
- Full L2 order book depth — See the complete bid/ask stack, not just top-of-book. Essential for gauging real liquidity before placing larger orders, spotting absorption, and reading order flow.
- Real-time funding rates — New endpoint and agent tool for accessing funding data. Build strategies that trade the funding rate itself, or use it as a filter to avoid entering positions with unfavorable funding.
- Proxy infrastructure — Webshare proxy support handles rate limits transparently during high-frequency operations. Your agents keep running even during heavy market activity.
- Historical positions endpoint — Leaderboard tracking with full position history. Study what top traders are doing and when.
Chart Improvements#
- Multiple new indicator types with multi-indicator support and custom parameters
- Better candle selection, improved indicator handling, and chart settings now persist to your account via Supabase
- Bug fixes: partial close sign on shorts, manual partial close, SL/TP direction on shorts, Hyperliquid 'k' token URLs, and S-market issues
Cross-Chain Bridging#
Moving assets between chains shouldn't require a PhD in bridge protocols. The new guided bridging wizard walks you through cross-chain transfers step by step: select your source chain, pick your destination, choose the asset, and the wizard handles bridge routing automatically. Token ID resolution bugs that were breaking certain transfer and swap pairs have also been fixed.
Portfolio: Rebuilt for Clarity#
The portfolio panel has been rebuilt from the ground up. Hyperliquid and GMX positions now render in a unified layout with consistent formatting — no more visual inconsistencies between venues. The panel is faster, cleaner, and designed to give you a complete picture of your holdings at a glance.
Additional improvements:
- Agent suggestions — When creating new positions, the system suggests relevant agents that could manage them
- Per-goal model selection — Goal cards now let you choose the AI model for each goal individually
- Cleaner production UI — Debug elements hidden, visual noise reduced
Account & Authentication#
- Google OAuth — Sign in with Google alongside existing providers. One click, no wallet required for initial setup.
- Enhanced profiles — Avatar and display name support so your identity carries across the platform.
The Invisible Work: Infrastructure & Scaling#
Most of what shipped this cycle isn't visible in the UI — and that's the point. Every feature you interact with sits on top of infrastructure, and when that infrastructure is slow or unreliable, everything feels broken. This was a major investment in the foundation.
Horizontal Scaling#
The platform can now run across multiple worker servers without conflicts:
- Distributed locking — When two servers try to process the same job, only one wins. No duplicate trades, no race conditions, no corrupted state.
- Multi-master pub/sub — Events broadcast correctly across all server instances. Every connected client gets real-time updates regardless of which server they're connected to.
- WebSocket connection pooling — More efficient real-time connections across the server fleet, reducing memory overhead and improving reconnection reliability.
Caching & Database#
Aggressive caching now covers candles, funding rates, orders, builder fees, and user tiers — dramatically reducing database and API load. Dozens of commits optimized database calls via join queries, batch operations, and smarter select clauses. The practical impact: faster page loads, faster agent responses, and significantly more headroom before we need to think about scaling again.
Monitoring & Stability#
- Sentry integration — Backend error tracking catches issues before they cascade into user-facing failures
- Crash prevention — Fixes for edge cases including missing positions, undefined results, and connection timeouts
- HL API proxy support — Rate limiting handled transparently with Webshare proxies and an API weight meter
Credits System#
Credits are now checked before running technical analysis, goals, and other paid operations — not after. No more operations failing mid-execution because your balance hit zero. The CDX token requirement for credit claims has been removed, and claim logic has been refined for zero-balance users to ensure a smooth experience.
Also Shipped#
- YouTube scraper — New daemon and tool for scraping YouTube content, expanding data sources available to agents
- Supabase realtime optimized — Frontend properly deduplicates and cleans up subscription channels, cutting unnecessary connections
- Feature updates modal — Updated to v1.3.0 with the latest release notes
This cycle represents hundreds of commits across the stack — but the throughline is clear: smarter agents, deeper data, and infrastructure that disappears so you can focus on trading. The next cycle is already underway with expanded ICT indicators, cross-agent communication, and the agent marketplace. We build in the open and shape the roadmap from your feedback. — CONCLAVE Inc.
