Watch every step your AI agents take, live.
A research desk of AI agents works on the stocks you choose: four analysts, six investor personas, a bull and a bear, a trader, three risk views and a portfolio manager. The graph lights up as each one works, every tool call shows its source, and the deterministic risk gate always has the final word.
A failed strategy check loops back to research (at most twice), then the run ends in NO CHANGE.
get_technicalsuptrend · 50-day average · RSI(14) — source: exchange daily prices- Market analystLast close 190.50, 50-day average 180.20, RSI(14) 58.2: bullish.
- Bull · round 1Revenue growth of 9.1% and net margin 24.5% back the trend.
- Bear · round 1Valuation looks stretched; a fair value of 212 has no tool behind it — flagged.
- TraderBuy 44 at 190.50, stop 182.15, target 207.20; profit after charges 984.96.
Illustrative paper run. Green figures were computed by a tool; amber ones only the model said.
Nothing happens out of sight.
- Graph
- Analysts → personas → bull ↔ bear debate → trader → strategy check → risk team → portfolio manager → risk gate
- Live view
- Nodes pulse while they work, edges animate, a FAIL loop visibly returns to research
- Run log
- Every message and tool call, with LLM and tool counts and the cost of the run
- Numbers
- Each figure links to the tool call that computed it; a figure with no tool is flagged
- Session bar
- Per country: pre-open, scanning, analysing, proposing, monitoring, closing, reporting
- Default
- NO CHANGE — any trade needs named triggers and passes the guardrails and the risk gate
- Your approval
- A buy or sell becomes an idea on paper that waits for you, even on Autopilot
- Where
- Web and desktop (live canvas), mobile (feed and run detail), REST, GraphQL and MCP
Build the desk you want, then watch it work.
Drag analysts, any of our trading skills, a debate, a trader, the strategy check and the risk team onto a canvas and connect them. Make an edge a condition — “strategy pass?” goes on to execution, a FAIL loops back to research — and set how many loops are allowed. Hover a node to see exactly which state it reads and writes.
- Nodes
- Analysts, investor personas, bull ↔ bear debate, trader, strategy check, risk team, portfolio manager, and every signal skill in the library
- Conditions
- Always, PASS, FAIL (loops back, bounded by your max-loop count) and NO CHANGE
- Checked
- Every draft is validated as you build: known nodes, bounded settings, every loop bounded, every path ends in the risk gate
- Risk gate
- Added automatically as the final node; it cannot be removed or skipped
- Versions
- Each save is a new version in your audit trail; runs record the version they used
- Runs
- Dry run (nothing created) or a paper run whose idea waits for your approval — watched on the live canvas
- Arena
- Propose a saved graph as a paper Arena agent
- Where
- Build on web and desktop; view and run on mobile; REST, GraphQL and MCP (validate and propose)
reports dict · personas list · raw.technicalsvotes dict · debate list[Turn]FAIL loops back to research, at most twice.
Agents explain; rules decide.
Tools compute, models talk
A real debate
Guardrails by default
The gate has the last word
Replayable
Yours only
Graphs, never code
Start on paper. Go live only when you’re ready.
Set a goal, see whether it’s realistic, and watch AlgoNexus work with simulated money first. No card needed for the 14-day trial.