Skip to content

Releases

November 2, 2026

More timely Crazy Mode performance updates, fewer interrupted run checks, and more reliable trade history and market-data requests.

  • More timely Crazy Mode updates. A slow account no longer holds up performance updates for other runs, reducing stale chart observations and delays in checking profit and loss exits. Temporary failures are retried sooner.
  • More reliable Crazy Mode run checks. Fixed a conflict between performance updates and run checks that could interrupt a check and delay it until the next attempt.
  • Consistent Crazy Mode P&L during fills. If a trade fills while the account is being read, Crazy Mode retries the observation instead of recording a point that could count the same price move twice.
  • Fewer failed market-data requests. Candle history, candle statistics, technical analysis, and market research now accept either assets or symbols for the asset list, avoiding failures caused by an agent using the other name.

November 2, 2026

Easier Crazy Mode starts, recorded run highs and lows, faster live-trade updates, and more accurate performance figures.

  • Crazy Mode highs and lows. See the highest and lowest recorded P&L, return percentage, and time in current-run, completion, and History details. These values come from periodic observations and can miss moves between them.
  • Start Crazy Mode from a paused agent. Starting a run now resumes the agent automatically. If the start fails, it stays paused. This also prevents an overdue normal decision from slipping in ahead of the run’s first decision.
  • More faithful Crazy Mode replays. Shared replays now use the run’s recorded performance and trade events, preserving gaps in the data and the saved final result. You can also share an active run’s progress so far.
  • Crazy Mode profiles on leaderboards. Agents in an active run now show their Guarded, Sprint, or Runner profile through their avatar’s gear.
  • Faster live-trade updates. Newly detected Hyperliquid trades prompt an earlier trade-history update, including take profits, stop losses, liquidations, and trades placed directly on Hyperliquid. Eligible agent order fills also prompt a follow-up decision sooner, instead of waiting for the next scheduled order check.
  • Better recovery from model outages. When a trading model repeatedly fails across agents, new turns for that model wait while DXAP checks for recovery. Other models can continue, and pending turns resume when the affected model responds successfully.
  • Correct funding in live P&L. Funding paid now reduces P&L and funding received increases it consistently in new live performance records and agent context. Historical performance records are unchanged.
  • Fewer misleading performance dips. Live chart points now use the time the account was read, preventing trades completed during a refresh from being left out of that point’s realized P&L.
  • More accurate returns after deposits and transfers. Performance updates check for newly arrived cash flows when account equity changes unexpectedly, reducing delays before percentage returns reflect the updated capital.
  • More consistent chart markers. Live performance charts and the Trades tab now share the same trade records, keeping trade details and agent labels aligned across both views.
  • Trading continues through performance-update failures. A temporary failure to refresh performance no longer stops agent processing. Failed updates are retried automatically.
  • Fewer order-format rejections. Fixed valid order prices and amounts being rejected when a model added an extra layer of quotation marks.

October 29, 2026

A shared chat thread for each Crazy Mode run, recorded replay data, and a paper-funding fix.

  • Crazy Mode run conversations. While you view a run, your agent posts commentary about its decisions and performance into that run’s own chat. Reply in the same thread to discuss what is happening. Your replies take priority over automatic commentary, and the conversation remains readable after the run ends.
  • Recorded Crazy Mode replay data. The API now lets you retrieve your completed run’s recorded performance, saved final result, and selected agent trade events for replay creation. Access is limited to the run’s owner, and incomplete observations remain marked as gaps.
  • More room for Crazy Mode positions. New runs can hold up to three symbols at once, sharing the same 95% allocation across all positions. Existing runs keep the profile they started with. See Crazy Mode.
  • Paper funding for kPEPE, kBONK, and kSHIB. Fixed a market-name lookup issue that could leave paper positions in these markets without funding settlements.

October 29, 2026

Updated Crazy Mode strategies, separate memory for each run, and fewer avoidable order rejections.

  • Updated Crazy Mode profiles. Guarded and Sprint can now go long or short and move on from stagnant trades. Runner allows up to six entries in any 24 hours. See Crazy Mode.
  • Fresh memory for each run. Crazy Mode keeps its trading notes separate from normal trading and other runs, and skips group chats.
  • Clearer profit and loss figures. Crazy Mode agents now receive precomputed USD amounts alongside percentage returns, reducing the calculations they need to do themselves.
  • Crazy Mode on leaderboards. See which agents currently have Crazy Mode active.
  • Fewer rejected orders. Agents automatically adjust leverage and small order sizes to meet exchange requirements within your trading limits.
  • Automatic price rounding. Prices with too many decimal places are rounded instead of rejected, with the adjusted price shown in the order details.
  • Correct trade history. Replacing or canceling a stop loss or take profit no longer makes a filled trade appear canceled.
  • Orders near trading limits. Fixed an issue that incorrectly rejected some live market orders within your limits.
  • More reliable technical analysis. Fixed requests failing because the agent used a different name for the asset list.

October 23, 2026

Individual candles for agents, consistent indicator values, invites for funded live agents, and more reliable order protection.

  • Candle history for agents. Agents with Technical Analysis enabled, agent chat, and market research can read individual completed OHLCV candles for up to 10 symbols per request. Use them for swing highs and lows, wick patterns, and candle-by-candle comparisons. Request up to 500 candles per symbol, with lower limits on some timeframes, such as 360 for 4h and 365 for 1d. Candles are returned oldest first, and any period without data is marked rather than skipped. See Candle history.
  • Two invites when you fund a live agent. If you don’t already have a referral code, funding the account of a live Hyperliquid agent unlocks two invites and sends you an inbox notification. Accounts funded after you set up the agent qualify after their next balance update.
  • Consistent indicator values. Indicators are calculated with the same history whether an agent checks them during a live turn or for an earlier time, so a past reading matches what the agent saw then. Technical-analysis results show each requested indicator’s previous and latest completed-candle values, with candle timestamps and calculation coverage. Use candle history for individual candles and candle statistics for aggregate windows.
  • One referral code per account. If you had more than one referral code, they have been combined into a single code that keeps their remaining invites. Claimed invites and the rewards they earn are unchanged. Links using your other codes no longer work, so share the code shown in Rewards.
  • Shorter referral codes. New codes use the XXX-XXX-XXX format and leave out easily confused characters such as O, 0, I, and 1. Existing codes are unchanged.
  • API keys without an expiry. Choosing Never as an API key’s expiry no longer fails after you sign, in agent settings or when connecting an external agent.
  • Take profits and stop losses after interrupted market entries. If confirmation of a market entry with attached TP/SL was interrupted, the protection is now placed once the complete fill is confirmed, sized to the filled quantity and the current position. It is skipped if the position has since closed or reversed, or its protection was changed. Previously, the position could be left without its requested TP/SL.
  • Closes no longer blocked by lost confirmations. A market order whose confirmation was lost could prevent later closes on that symbol. These orders are now matched with the exchange’s records, without being resubmitted.
  • Rejected orders release rolling entry limits. Orders that the exchange definitively rejects no longer count toward Total Entries per 24h or Total Entry Notional per 24h. Open, filled, and unconfirmed orders still count.
  • market_technical_analysis returns indicator values only. Remove the lookback input from your requests; requests that include it are rejected. Results no longer include recent candles. Request the indicators and periods you need explicitly; omitting them returns the default set.

October 17, 2026

  • Missed points and referral rewards restored. Eligible fills from live limit orders and take-profit and stop-loss orders placed through DXAP have been credited with the volume points and referral rewards they missed, based on each recorded fill. Rewards already awarded were not changed, and trades placed outside DXAP are not included.

October 17, 2026

  • Points and referral rewards for all live DXAP orders. Fills from live limit orders and take-profit and stop-loss orders now earn volume points and referral rewards in the same way as market orders, including partial fills and orders that were edited or replaced.

October 16, 2026

  • Optional reasons for external-agent trades. External agents can omit reason or send null when submitting, closing, or managing trades through REST or MCP. Reasons that are supplied are still recorded.

October 15, 2026

  • Full-length baselines for technical analysis. Candle-statistics checks can now use the full supported baseline plus a separate comparison candle. For example, a seven-day baseline of 672 completed 15-minute candles can be compared with the next completed candle without being rejected for exceeding the history limit. Existing timeframe limits and the 1,000-candle total cap still apply.

October 15, 2026

  • One MCP connection endpoint. Account tools and external-agent tools now share the public /mcp endpoint. The tools available to each connection depend on its API key’s permissions and bound agent and account.
  • Update external-agent MCP connections. Replace /external/mcp with /mcp on your public MCP server. The old endpoint has been removed.

October 15, 2026

  • Bring your own trading agent. Connect an external agent to a paper or live Hyperliquid account through REST or MCP. Your software controls its decisions, memory, and schedule while DXAP executes supported trades and tracks portfolio performance and activity. External agents do not use DXAP-hosted chat, schedules, or triggers.
  • Control external-agent access. Create API credentials tied to a specific external agent and account, with separate permissions for reading context, market data, and history, submitting trades, and managing orders. External agents can place market and limit orders, reduce or close positions, edit or cancel eligible orders, and manage take-profit and stop-loss orders. Existing trading limits and live-account approval requirements still apply.
  • Review and approve trades in agent chat. Hosted agents can prepare trade cards for market or limit entries, partial or full closes, limit-order edits or cancellations, and position take-profit and stop-loss orders. Review the proposed terms, then choose Approve to submit or dismiss the proposal. Chat submits the stored terms only after your approval and checks current account permissions and order conditions before execution.
  • Carry approved trade context into later decisions. Approved chat trades and any rationale you provide are available to subsequent autonomous turns. Preparing a trade card does not pause autonomous trading.

October 11, 2026

  • Fewer unnecessary agent turns after order updates. Order cancellations and rejections no longer trigger a follow-up trading turn, including automatic TP/SL cancellations after a position closes. Follow-ups are limited to newly filled quantities on orders attributable to that agent, avoiding extra turns from unrelated account activity or fills already reported at submission. Delayed and partial fills can still trigger reassessment.

October 9, 2026

  • Better order context in agent chat. Chat now retains open-order information and can look up the latest recorded state of historical orders, including filled or canceled orders and their linked take-profit, stop-loss, or replacement orders. Observation times help distinguish the latest order state from what was known when the original trading decision was made.
  • More reliable Hyperliquid order confirmations. Accepted limit orders, replacement orders, and take-profit or stop-loss orders no longer appear as unknown merely because an immediate status lookup fails. Partially filled limit orders retain their remaining quantity, and eligible accepted orders retain their edit and cancel options.

October 9, 2026

  • Limit orders for trading agents. Agents can now place resting limit orders, review account-wide open orders, and edit or cancel eligible orders. Unfilled entry orders count toward the open-position cap by symbol.
  • Automatic take profits and stop losses. Agents can attach TP/SL when submitting a market or limit entry in the same request, or update protection on an existing position with a stop and up to four take-profit targets. Market-entry exits cover the confirmed filled quantity; limit-entry exits wait for the full entry fill. Use orders for automatic price-based exits and triggers for conditions that need agent reassessment. See Orders for examples.
  • Settings change history. Open Change history beside Edit to browse saved agent configurations, compare old and new values, and see highlighted strategy edits.
  • Restore saved settings. Preview an earlier configuration against your current settings before restoring its strategy and settings. Restoring creates a new history entry and preserves earlier versions. Account, schedule, running status, visibility, notifications, and memory stay as they are; trades are not undone.
  • Lifetime totals and summary sorting. See each agent’s lifetime trading volume and trade count separately from trades in the activity window. Sort agents in either direction by performance, lifetime totals, activity, name, status, open positions, active triggers, or next turn. Unavailable values appear last.
  • Remove chats from Recent chats. Delete a conversation from your history after stopping or finishing any active response. This archives the conversation; it does not erase stored transcripts, change agent instructions, or clear saved memory.
  • Lifetime trading totals in the overview API. Agent entries now include total trade count and trading volume in USD across the agent’s attributed accounts. Totals are independent of the chart range and selected account. Volume includes opening and closing fills, and totals may lag the newest trades until the next refresh.
  • Easier browsing of long chats. Conversations open with recent messages and load older history as you scroll up, preserving your reading position. Charts load as you approach them, and other tool details open on demand. Failed history and chart loads offer retry controls.
  • More reliable agent turn completion. Fixed a response-format issue that could mark an otherwise valid trading turn as failed when the model returned its final summary as text.

October 6, 2026

  • More compute time for complex strategies. We’re giving your agents more time to work through demanding analysis and detailed strategy decisions. The time allowed for an individual model request has doubled from 5 to 10 minutes, giving longer-running reasoning more room to finish. Overall chat and trading-turn time limits still apply.

October 6, 2026

More reliable agent conversations and trade history, plus clearer identity information for public leaderboards.

  • Leaderboard wallet and ENS support. Return and Volume leaderboard data now includes the agent owner’s wallet address. Return, Volume, and Points leaderboards also support ENS names when available, helping identify participants. Leaderboard results remain available when a name cannot be resolved.
  • Chat continues through connection drops. Once your message is accepted, refreshing the page or losing your connection no longer cancels the agent’s response. Completed answers are saved to the conversation.
  • More reliable trade history. Improved synchronization catches Hyperliquid trade fills that arrive late, reducing gaps in recorded trading activity without duplicating trades.
  • Group chats with more supporting sources. Fixed failures that could prevent agents from reading group chats with many evidence sources.

October 3, 2026

More precise technical checks directly in agent chat, building on the expanded TA toolkit in 0.9.50.

  • Direct candle statistics in agent chat. Ask for completed-candle trend, breakout, range, or volume measurements directly in your selected agent’s chat. Matching checks can cover up to 20 symbols per request, preserving each symbol’s values, window boundaries, and data coverage.
  • Focused technical lookups. Chat uses candle statistics for exact window calculations and its existing indicator tools for measurements such as ATR(14). Market research continues to handle broader synthesis, news, and catalysts.
  • Aligned trend and breakout checks. Updated guidance pairs the latest completed trend window with the preceding breakout baseline and uses a shared observation time for related indicator requests. For example, compare a seven-day change over 42 completed 4h candles with a breakout above the preceding 42 candles’ highest high, then request 4h ATR(14).

These chat lookups are read-only. Strategy and settings changes retain the proposal and approval flow, and autonomous technical checks still require Technical Analysis to be enabled. A chat result does not replace fresh evidence on a later trading turn.

October 3, 2026

Expanded technical analysis, persistent agent-chat context, and additional execution controls.

  • Agent-scoped chat memory. When enabled, saved discussion preferences, goals, and user-adopted market theses carry into new conversations with the same agent. Ask the agent to save, correct, or forget a detail. Background processing can also capture durable context from newly completed exchanges. Chat memory remains separate from trading notes and does not change strategy or settings.
  • Maximum Open Positions. Set an optional cap on distinct open symbols on the selected account. Manual positions, positions opened by other agents on that account, and pending entries count toward the cap. Increasing an existing position does not consume another slot. Reductions and closes remain available, and lowering the cap does not force closures.
  • Re-entry Cooldown. Set an optional waiting period, in minutes, after a symbol is fully closed. The rule covers subsequent long and short entries, including after manual closes and liquidations. Partial reductions do not start the timer. Changing the duration applies to the last recorded full close.
  • More context in agent chat. Chat can consult the selected agent’s trading notes and request broader market research. Reading trading notes is available independently of chat-memory capture.

Position caps and cooldowns govern agent entries; manual trading retains its existing permissions. Both settings accept positive whole numbers, and leaving them unset disables the respective limit. See Agent settings and Chat and instructions for configuration and approval flows.

  • More powerful technical analysis. Trading agents with Technical Analysis enabled can access indicator and completed-candle measurements directly, building on their existing TA and research capabilities. Direct tools handle precise numerical checks; market research continues to provide broader analysis, news, and catalyst context.
  • Trend and breakout measurements together. Each candle-statistics request returns the latest trend window, preceding baseline windows, and the latest completed comparison candle. For a seven-day trend plus a 42-candle breakout on 4h candles, the trend window includes the latest candle and the breakout baseline excludes it. The seven-day measurement ends at the latest completed candle, rather than the current moment.
  • Larger candle scans. Check up to 20 symbols per candle-statistics request when timeframe, window size, window count, and observation time match. Compact tables retain numerical precision, timestamps, coverage, and comparisons, with shared definitions reported once. This does not automatically make every turn scan every market.
  • Explicit indicator selection. Updated guidance helps agents request indicators and periods such as 4h ATR(14), including for volatility-based stop calculations. ATR must be requested explicitly; it is not part of the default indicator set.
  • Clearer strategy reviews. Chat distinguishes evidence recorded under earlier configurations from behavior under updated settings and identifies when relevant history is insufficient to assess a change.
  • More reliable candle requests. Candle-statistics calls now accept the string-encoded symbol arrays some models produce, avoiding unnecessary validation failures.
  • Actionable data-quality reporting. Routine unverified-integrity warnings no longer appear in agent-facing candle results. Actual missing, incomplete, or invalid candles still make affected measurements unavailable, and required unavailable evidence prevents the dependent entry.
  • Clearer leverage rejections. Opening orders require a known exchange maximum leverage. Requests above that limit are rejected before submission with the requested and allowed leverage, instead of a generic leverage-update failure. Missing leverage metadata is reported as unavailable. Closing orders retain their existing path.

For callers of market_candle_statistics:

  • Remove the compare_latest argument. Responses always include latest_window, preceding windows, and comparison_candle.
  • Read the full evidence from the compact text tables. Structured content now contains only status, format, and result/error counts; it no longer contains the full results array.
  • Candle statistics accepts up to 20 symbols. The separate indicator tool, market_technical_analysis, retains its 10-symbol limit.

October 1, 2026

  • Public agent positions. Read-only access to a public agent’s current positions, including symbol, side, size, entry and mark prices, leverage, and unrealized profit or loss. Position results support cursor pagination.
  • Public performance history. Read-only chart data for a public agent, with 1h, 24h, 7d, 30d, and 90d ranges and validated chart intervals.
  • Visibility protection. These public responses require an explicitly public, available agent and omit private account identifiers, account names, and private activity details. Private, deleted, or unavailable agents return not found.

Public API routes:

GET /api/v1/public/agents/{agentID}/positions
GET /api/v1/public/agents/{agentID}/performance/history

These routes require no authentication and return Cache-Control: no-store. Existing authenticated routes are unchanged.

October 1, 2026

  • Clearer position context for trading agents. Positions are supplied as labeled records, making fields such as side, size, entry price, margin, and leverage explicit while preserving numerical precision. Unavailable values remain marked as unknown.
  • More focused trading memory. Resolved and other explicitly completed notes are excluded from active trading context while their stored history is retained. Active and ambiguous notes remain available to the agent.
  • Less repetition during memory updates. Trading-memory saves and deletes return concise confirmations and focused correction details, reducing repeated context during a turn.