Skip to content

Hummingbot v2.16.0 Release Notes

Released on July 29, 2026

Repository Description GitHub Release DockerHub Release
Hummingbot Client Core Client v2.16.0 changelog version-2.16.0
Gateway DEX Middleware v2.16.0 changelog version-2.16.0

Note

Other Hummingbot repositories such as Condor and Hummingbot-API follow a continuous deployment model without fixed version releases. Use the main branch for these repositories.

How to Install / Update

Agent Builders Cup Hackathon

The Agent Builders Cup is a agentic strategy hackathon run by Botcamp, our affiliated education company. Build a Hummingbot V2 Controller or Condor Agent, apply to one of the exchange sponsor teams — Orca, Derive, Gate, XRPL, Bitget, and Meteora — and compete for a $20,000 total prize pool. Each sponsor selects two agents, and the agents compete in a 48-hour live trading competition.

The build window opens August 1, 2026 and runs through August 31, followed by judging in September, a livestreamed finals on October 1–2, and winners announced October 7 at our Token2049 side event in Singapore. See botcamp.xyz/hackathons for the full schedule, rules, workshops, and team sponsors.

Register

🐦 What's New - Hummingbot Client

💻 hbot CLI

Added hbot, a non-interactive command-line interface for running, controlling, and monitoring Hummingbot without the interactive TUI. Every command emits compact Markdown output with stable exit codes, and run/observe support --json for machine-readable output—enabling AI agents and humans alike to run Hummingbot from scripts, CI, and agent-driven workflows. It drives the same HummingbotApplication and TradingCore engine as the classic client through a headless host, with no MQTT broker required.

See the hbot CLI documentation for install instructions, commands, and examples.

PR: #8337

🔌 New Connectors (Gemini & Lambdaplex)

  • Gemini – Spot connector for the regulated cryptocurrency exchange. Trade via the Gemini Fast API (WebSocket) and REST v1 API with full ExchangePyBase support—limit and market order placement, cancellation, balance tracking, and real-time order and trade updates over WebSocket (API docs). PR: #8027 | Thanks to erickuhn19! 🙏
  • Lambdaplex – Spot connector for the Hedera-native onchain exchange with an offchain matching engine and onchain settlement. Trade via REST and WebSocket with full ExchangePyBase support—limit and market order placement, cancellation, balance tracking, and real-time order book and trade updates; authenticate with an API key and unencrypted private key (API docs). PR: #7889 | Thanks to petioptrv! 🙏

🛠️ Connector Updates

  • Hyperliquid: Fixed perpetual order price quantization to exchange tick size (#8356). Thanks to chevoisiatesalvati! 🙏
  • Binance Perpetual: Fixed order book diff sequencing and rate-limit handling (#8291)
  • KuCoin Perpetual: Fixed trade stream WebSocket topic and timestamp field (#8313). Thanks to ipezygj! 🙏
  • Backpack: Improved balance, leverage reporting, and rate oracle source (#8323)
  • Bitget: Added LIMIT_MAKER (post-only) order support to spot and perpetual connectors (#8335). Thanks to isreallee82! 🙏
  • Removed connectors: Removed Cube, AscendEx, and Vertex connectors (#8376)

💧 LP Executor Fixes & Improvements

Several fixes improve LP Executor reliability for live CLMM liquidity provision on Gateway DEXs—especially full position unwinds, accurate swap fill reporting, and graceful shutdown.

  • keep_position=False now fully unwinds — the close-out swap sells the entire withdrawn base balance instead of the delta against the initial deposit, preventing silent wallet residue after an EARLY_STOP.
  • Gateway swaps report realized fills — swap results read confirmed on-chain amountIn/amountOut instead of the requested amount, so follow-on LP opens don't fail when slippage reduces the wallet balance.
  • Shutdown preserves position exposure — executors that can't finish closing before the shutdown deadline are force-stopped into POSITION_HOLD records; the orchestrator no longer re-calls early_stop on executors already shutting down, and the wait budget extends while executors make progress. StrategyV2Base.on_stop also restores market registrations before executor close so in-flight orders can still execute.
  • net_pnl_pct consistency — returns a ratio like other executors instead of whole percentage points.

PRs: #8310 | #8389 | #8391 | Thanks to mlguys and erickuhn19! 🙏

⚙️ Strategy V2 Executor Fixes

  • #8362 – Fixed OrderExecutor close orders to set position_close=True on perpetual connectors Thanks to lgrawet! 🙏

Other Updates

  • #8304 – Improved candles feed initialization and historical candle fetching
  • #8338 – Fixed candle requests to always span at least one interval Thanks to matthiola0! 🙏
  • #8311 – Resolved race condition in account readiness for WebSocket assistants
  • #8328 – Updated rate oracle to treat USD as USDT-equivalent for balance conversions
  • #8348 – Fixed global stop-loss spam after position close in pmm_mister Thanks to ybapat! 🙏
  • #8377 – Fixed Gateway order sizing for mint-address tokens and skipped BudgetChecker for Gateway swaps

🌉 What's New - Gateway

🔗 New Chains (Robinhood Chain & Unichain)

Gateway adds support for Robinhood Chain (robinhoodchain, chainID 4663) and Unichain (chainID 130), with Uniswap router, AMM, and CLMM wired on Robinhood Chain mainnet using official deployment addresses.

See the Uniswap connector page for how to connect to these networks, and Chains for general chain setup.

PR: #665

🔀 New Routers (Titan, DFlow, OKX DEX)

Gateway adds Titan, DFlow, and OKX DEX as Solana router connectors alongside Jupiter, and consolidates Solana transaction signing through a single signing chokepoint.

See Connectors for setup and usage.

PR: #665

🦄 Uniswap Improvements

  • Uniswap v4 routing — Routing now includes v4 pools on chains in the Uniswap SDK's v4-supported list (mainnet, Arbitrum, Optimism, Polygon, Base, BSC, Avalanche, Celo, Unichain, and more); other chains continue to route across v2 + v3 only. No API or approval changes: v4 swaps execute through the existing Universal Router + Permit2 flow — the router simply considers v4 pools and picks whichever route prices best. On Unichain, on-chain quote batching is reduced from 80 quoter calls to 10 per multicall so public RPCs accept quote requests reliably.
  • Permit2 allowances no longer expire by default — Permit2 grants from /approve (used by uniswap/router) were previously hardcoded to expire after 48 hours, forcing bots to re-approve every trading pair every 2 days. A new optional permit2ExpirationSeconds setting in the Ethereum chain config controls the expiration; when unset, grants never expire, matching plain ERC20 approval semantics on other connectors. Allowances remain amount-scoped and revocable.

PR: #665

🔒 Wallet Security

Hardened Gateway wallet security for network-reachable deployments: at-rest KDF for wallet keys, loopback bind by default, opt-in API authentication, and rate-limiting. Controls engage only for non-loopback traffic, so local and Docker setups are unaffected. Hummingbot API now auto-generates shared mTLS certificates on first start to match.

PRs: gateway #659 | hummingbot-api #185

Other Updates

  • #656 – Added retry for read RPC calls on rate-limit errors for Solana and Ethereum Thanks to mlguys! 🙏

🦅 What's New - Condor and Hummingbot API

🧠 Agent Memory & Skills

Condor agents now support persistent user memory and reusable skill playbooks—markdown guides the agent retrieves by relevance and injects into chat context. Memory and skills are scoped per assistant, exposed via MCP tools (manage_memory, manage_skill), and manageable with the new /memory command.

See Condor for documentation.

PR: #135

🤖 OpenRouter Model Picker & Advisor

Building on OpenRouter support from v2.15.1, Condor's web dashboard now has full model picker parity with Telegram /agent—including OpenRouter, Ollama, and LM Studio—so you can browse and select models from the live catalog in the browser. A new agent model-selection advisor inspects what's configured (cloud API keys, local LLMs) and recommends a model suited to each agent's role, steering you toward local models or free OpenRouter options when appropriate to reduce token usage and costs.

How to use:

  1. Create an API key at openrouter.ai/keys (optional for browsing; required to run OpenRouter models)
  2. Add OPENROUTER_API_KEY=sk-or-... to your Condor .env file, or run Ollama/LM Studio with a tool-capable local model
  3. In the web dashboard, start a new chat and pick from the unified agent/model dropdown—OpenRouter models show live pricing, including $0 free options
  4. When building agents, the advisor recommends a model based on your setup and the agent's complexity

PR: #161

📊 Interactive Reports

Condor reports are now self-contained interactive trading reports with shared datasets, filters, computed metrics, linked charts, drilldowns, searchable tables, CSV export, and optional auto-refresh. Generated HTML embeds its data and browser assets, so reports work in the Condor dashboard or as downloaded offline files. A new Report Component Gallery demonstrates the full ReportBuilder surface with simulated data.

See Condor for documentation.

PR: #157 | Thanks to mlguys! 🙏

Other Updates

  • #140 – Revised README for Hummingbot API naming and Tailscale security recommendations
  • #143 – Added DELEGATE execution mode for fire-and-forget background agent tasks
  • #146 – Fixed agent drawdown calculation to measure from exposure instead of peak profit Thanks to mlguys! 🙏
  • #159 – Fixed Codex ACP package integration to use maintained @agentclientprotocol/codex-acp bridge Thanks to mlguys! 🙏
  • #185 – Secured Gateway with auto-generated shared mTLS certs
  • #190 – Replaced RateOracle with in-house ticker pool and GeckoTerminal DEX pricing
  • #193 – Updated Gateway client to use unified /trading endpoints and stop masking Gateway errors
  • #182 – Added Makefile reset target for cleaning up Docker and credentials
  • #184 – Enhanced README with Tailscale setup and API access details
  • #192 – Fixed rate finder import after hummingbot token normalizer rename