Gogi

Link an algorithm to a trading instance with capital allocation and start live trading. algo_id is the saved algo id (from create_algo / list_my_algos), not LinkedAlgo id. instance_id is from create_instance / list_instances. ALWAYS ask the user for capital_allocation — never invent a dollar amount. Capital is from broker buying power; the cumulative sum across this agent's live algos (allocated AI Trading Power) must stay ≤ policy max_total_capital. First call with confirm_allocation=false; server returns allocation_requires_confirm with broker_equity, allocation_pct_of_equity, and warning_level (high≥50%, critical≥90%). Show those to the user with risk settings; after explicit OK retry with confirm_allocation=true (acknowledge_capital_risk still accepted as alias). capital > equity is hard-blocked. Optional risk_preset override: conservative|moderate|aggressive|custom. On success check engine_started (or utml_started), is_live, symbols, timeframe, and capital vs equity.

Chart patterns, S/R, Fibonacci, and indicators. Default overlay=patterns. Summarize steps performed when presenting results. Use get_market_data for quotes.

Cancel a pending trade before in-app confirmation. This cannot cancel an already executed, expired, or previously resolved trade.

Close open positions on a Connections account (all asset classes, all brokers). ONE call closes every matching position or ticket: omit symbol to flatten the account, or pass symbol to close every ticket for it (e.g. all TradeLocker XAUUSD tickets). Never loop over tickets one at a time to close a stack. If the agent requires confirmation (HITL) and several positions match, the first call returns status=confirmation_required with positions_to_close: show them, then repeat the call with confirm_close_all=true. Fully autonomous agents (AITL) close in the first call. Use position_id only when the user wants ONE specific ticket. May return pending_confirmation when HITL is on.

Create a trading algorithm strategy with full technical parameters, multi-condition entry/exit blocks, and risk settings. Supports all indicators (SMA, EMA, WMA, LWMA, SMMA, VWMA, DEMA, TEMA, HMA, KAMA, RSI, MACD, Bollinger Bands, VWAP, ADX, STOCH, ATR, CCI, WILLR, MFI, OBV, ROC, KC, DC, ICHIMOKU, PSAR, SUPERTREND, Price, Value). You can send the full strategy block EITHER inside strategy_params (object or JSON string) OR via flat top-level fields (strategy_type, indicators, buy_conditions, sell_conditions, buy_exit_conditions, sell_exit_conditions, stop_loss_percentage, take_profit_percentage, risk_percentage, max_drawdown, trailing_stop_enabled, trailing_stop_amount, allow_pyramiding, max_pyramid_positions, close_on_bias_flip, trade_direction). Use the strategy the user (or this conversation) specified. If indicators/entry conditions were not specified, ASK — do NOT invent EMA 9/21 or any favorite indicator pair. Only pick a concrete template when the user explicitly invites you (e.g. 'you pick', 'surprise me'); then prefer diverse types such as RSI(14), MACD, SMA 50/200, or EMA 12/26 — never always the same MA pair.

Create algo + instance + start live trading on broker_account_id (account_ref from list_brokers; prefer paper/demo). Use the strategy the user (or this conversation) specified. If indicators/entry conditions were not specified, ASK — do NOT invent EMA 9/21 or any favorite indicator pair. Only pick a concrete template when the user explicitly invites you (e.g. 'you pick', 'surprise me'); then prefer diverse types such as RSI(14), MACD, SMA 50/200, or EMA 12/26 — never always the same MA pair. Ask for capital_allocation; keep cumulative live capital ≤ policy max_total_capital. Prefer user risk numbers in strategy_params.risk; optional risk_preset otherwise. First call confirm_allocation=false; after user OK retry confirm_allocation=true. On success present Instance Setup Details.

Create a trading instance for broker_account_id (account_ref from list_brokers). Prefer paper/demo unless user chooses live. Default status pending; start_immediately=true only when user wants live now.

Deactivate and unlink an algorithm from a trading instance, stopping that linked strategy without deleting the saved algorithm. Use this before delete_algo; it is reversible with activate_algo, but inspect open positions first because deactivation does not automatically flatten broker exposure.

Archive a saved algorithm owned by the user. Requires confirm=true. An algorithm linked to a trading instance cannot be archived until it is deactivated, preventing accidental disruption or loss of execution history.

Permanently delete a trading instance and free a slot (max 10 per user). Stop ≠ delete. For live instances pass force=true. If open positions exist, the tool returns error_type=open_positions_require_confirm with open_positions — show them to the user, warn that delete will close all open positions, then retry with confirm_close_positions=true after explicit user OK.

Money-moving Kalshi/Polymarket order (may be irreversible). Policy and in-app confirmation still apply when enabled. Use search_predictions to find contracts when the ticker is unknown.

Money-moving broker trade (may place a real order). Optional instance_id for algo-tracked live instance; omit for on-demand on broker_account_id/default. If mcp_require_confirm, returns pending_confirmation — Confirm in Gogi Activity. Use close_positions to flatten rather than inventing a sell quantity. order_type limit/stop/stop_limit need limit_price and/or stop_price; supported types depend on the broker (Alpaca all; TradeLocker market/limit/stop; Coinbase, Kraken, OANDA, Schwab market/limit; others market) — unsupported types are refused, never sent as market. Pass a unique client_order_id so a retried call cannot place a second order. Forex/CFD quantity is in lots.

Retrieve one saved algorithm owned by the user by ID. Use list_my_algos first when the ID is unknown; use get_linked_algo for an instance-specific runtime copy.

Get details for a backtest run by run_id (must belong to this user's agent).

Free. This agent's Gogi credits today: used, limit, remaining, status (ok/low/exhausted), when they reset (user's local midnight), and how to add a +100 credit pack. Check before long or expensive runs; tell the user when low.

Query a subscribed marketplace data feed by feed_id. Pass a ticker in query (e.g. AAPL). Returns FMP-backed data for financials/insider/congress feeds, or a text preview for uploaded file feed ids.

Read a LinkedAlgo on a live instance owned by this agent: parameters, capital, symbols, mcp_require_confirm (null=inherit), and effective_confirm.

FMP quote + full technicals (EMAs, stoch, MACD, RSI, ADX) + chart structure (patterns, S/R, fib) for stocks/crypto/FX/commodities/indices. No broker required. Optional timeframe (default 1day). Use get_data_feed for marketplace SKUs; analyze_chart_structure for other overlays/lookbacks.

Check status of a pending MCP trade awaiting in-app Confirm (pending, executed, cancelled, expired, failed).

Return Gogi's trading/MCP risk disclosure (optional; also summarized in initialize). User controls permissions; no investment advice; losses are the user's.

Fetch this agent's global policy constraints plus per-live-instance HITL/AITL overrides (instance_overrides). mcp_require_confirm on policy is the default; LinkedAlgo overrides apply only to that live binding. Returns agent_id/agent_name/bound_key (the agent this key is bound to — compare with the agent the user has open in Gogi) and enabled_accounts (brokers execute_trade will actually allow).

Get account-level balances, equity, buying power, positions, and Web3 wallet holdings across connected brokers and wallets. Read each account's pnl_summary for equity, balance, unrealized_pnl and daily_pnl: a null value means the broker did not report it (say 'unknown', never assume 0 or break-even); partial_errors says which call failed. Accounts in equity_unavailable_for are excluded from the totals. Use get_positions when you only need normalized open exposure. Label each account by its display_label (broker · environment · •••••last5 of the broker's own account number); never call an account by broker_account_id, which is Gogi-internal. Each position has attribution: 'this_agent' or 'outside_this_agent'. Read policy_context before judging risk: the agent's policy (incl. max_daily_loss, measured on the agent's own realized P&L) does not apply to positions the user opened outside this agent — never call those a policy breach.

Get normalized current open positions and exposure across connected brokers. This is narrower than get_portfolio and does not replace account balances or wallet holdings. Label each account by its display_label (broker · environment · •••••last5 of the broker's own account number); never call an account by broker_account_id, which is Gogi-internal. Each position has attribution: 'this_agent' or 'outside_this_agent'. Read policy_context before judging risk: the agent's policy (incl. max_daily_loss, measured on the agent's own realized P&L) does not apply to positions the user opened outside this agent — never call those a policy breach.

Recent FILLS as the broker reports them: the broker's own side (buy/sell) for each order, fill price, quantity, and entry_or_exit from the broker's open/close flag. Use this — not activity notices — to establish what actually happened (e.g. whether a ticket was closed and in which direction). If journal_side_mismatch is true, Gogi's journal disagrees with the broker: report the broker's side. Label accounts by display_label.

List recent completed or running backtest runs for this agent across MCP, chat, voice, and API. Use get_backtest_run for one run's details; use run_backtest to start a new run.

List all connected broker accounts and wallets (not Alpaca-only). Each row has display_label (broker · paper|live · •••••last5) and integer id for broker_account_id. Prefer paper/demo; never invent IDs. If none connected: https://gogi.ai/voice-agent.

List all available data feeds (plan, marketplace subs, user uploads).

List trading instances (id, name, status, is_live, broker). Stopped/pending still count toward the max-10 slot limit until deleted.

List saved algorithms (algo_id, strategy, symbols, timeframe, archived). Use when the user names an algo rather than supplying its ID.

List pending trades for this agent awaiting in-app Confirm.

Log and persist a custom activity event. Writes a permanent, non-idempotent event record to Gogi's backend database (retained indefinitely) which is displayed to the user in the UI Activity Feed / agent timeline. Repeated calls with the same details will create duplicate timeline records.

WHEN TO USE: Streaming inner thoughts, reasoning steps, heartbeat signals, current status, or execution milestones to provide an audit trail of the agent's thinking process. WHEN NOT TO USE: Do NOT use this tool for reporting trades, orders, or portfolio/position queries. Standard action tools (e.g., execute_trade, close_positions, execute_prediction) automatically emit their own rich timeline events; logging them again is redundant and creates database spam.

Read a user-uploaded file owned by this agent. Returns metadata plus a UTF-8 text preview (capped); binary files return metadata only.

Run a policy-aware historical backtest for this agent. Results appear in the agent's Backtesting tab and Activity feed. Requires Backtesting Pro.

Search public prediction markets (Kalshi and Polymarket). No connected Kalshi/Polymarket broker account required. To place a bet, use execute_prediction with a connected account.

Start automated execution for a stopped/pending instance (can enable live trading). Does not replace activate_algo or create_instance.

Stop automated execution for a running instance (reversible via start_instance). Does not flatten positions — use close_positions when the user wants that.

Update a saved algorithm owned by the user (and sync any linked instances). Only supplied fields change. Accepts strategy updates via strategy_params OR flat top-level fields (indicators, buy_conditions, sell_conditions, buy_exit_conditions, sell_exit_conditions, stop_loss_percentage, take_profit_percentage, etc.).

Update an owned trading instance's name or runtime settings. This does not replace activate_algo or update_linked_algo; setting is_live=true can start live execution and requires explicit user intent.

Update this agent's LinkedAlgo on an instance (parameters and/or mcp_require_confirm). Changing mcp_require_confirm overrides global Agent Policy for THAT live instance only — first call returns loop_mode_requires_confirm with disclosure; after user Continue, retry with confirm_loop_mode=true. New instances inherit policy until changed.