Make yours

Themes, providers, workflows, and tools, installed with one command.

/
Agents & Providers69

Agents & Providers 69

Add, choose, configure, route, or coordinate who runs a thread.

Account Pool BalancedKeep coding across multiple accounts when one reaches its limit. Route Claude Code and Codex threads with quota-aware balancing, primary and reserve accounts, per-account caps, and session affinity. Connect Kimi, Z.AI, OpenCode Go, Cursor, and experimental Devin Local CLI; compatible external clients can use the proxy too. Bring your own provider accounts, subscriptions or API keys. Requires BB 0.43.1+ and Node/npm; disable BB's bundled Account Pooler because both register the same routes and command.Agents & ProvidersOXINew
Agent RetryRetry a failed turn instead of accepting it as the end of the work. It retries any failure a provider reports, including one with no error classification at all — a gateway error, a process that died mid-stream, a request refused at the door — which bb's own retry declines. The wait grows with each attempt, up to 50 by default, and can be scoped by project, provider, thread, or the half of the failure space where the provider never took the request. Every decision is recorded with the rule behind it, and `bb agent-retry log`, `explain` and `simulate` show a retry before or after it happens. Disable BB's bundled Provider Retry plugin: both queue retries for the same failed turn.Agents & ProvidersEuan GuoNew
Agent ToolsShow which MCP servers and skills each BB machine actually has, then bring them to one catalogue and one canon. A matrix marks present, missing, and drifted configs; new servers wait for adopt, keep-local, or hide. Sync adds what is missing and never deletes on its own. The Skills tab treats ~/.agents/skills as a git canon, with fan-out into CLI homes and a version archive. Writes back up each config first; loopback paths and local secrets stay on that machine. Optional MetaMCP listing needs that gateway's API key.Agents & ProvidersKirill Vechkasov37
Fast Switch AccountsSwitch to another Claude Code or Codex login when the current one runs out of quota, without leaving BB. See each saved account's remaining limits before you pick one, then start the next thread on it. Add an account through the provider's own browser sign-in. Saved logins stay in the macOS Keychain. Tokens go only to Anthropic and OpenAI. Cursor and other providers show limits only. Runs on macOS only. Codex switching needs the Codex CLI signed in with ChatGPT and the Xcode Command Line Tools.Agents & ProvidersYuriy EgorovNew
Command CodeRuns BB threads on any of Command Code's 70 coding models, so a bug has to survive reviewers on Claude, GPT, GLM, Kimi or DeepSeek without leaving BB. Shell commands, file reads and edits arrive as BB timeline cards with streamed output and diffs, and reasoning stays separate from the reply. Needs Command Code installed and signed in, plus the adapter this repository ships. Command Code's headless mode cannot prompt per tool call, so writes are one setting rather than a per-call approval.Agents & ProvidersBukhori Muhammad Aqid19
CrewRun a standing team of agents that message each other, hand off work and wait for you only when they need an answer. You describe the team once in a crew.yaml: each member gets a fixed address, its own provider and model, a role and its own worktree, and every member is a thread in the sidebar. Several crews in one project coordinate through their leads, BB Tasks and the base branch, with a step limit so they cannot loop forever. A project overview shows who works, who waits and every cross-crew message. Merge and rebase need the repository on the machine BB runs on.Agents & Providerssajov27
Desktop & Browser AgentLets your agents use the apps and browser tabs on your machine in the background, while you keep working. Agents read and operate native apps and Chrome, Edge or Brave windows through Cua Driver from Cua AI, which the settings page installs and checks on each machine. The optional Browser Bridge for bb extension lets them work in any tab of your signed-in browser, in an Agent tab group you can pause. Tools that run code or force-quit apps start off, and agents can be made to ask before clipboard, screen capture and file actions. Needs Cua Driver (free, MIT). Tested on macOS 14+; Windows 10/11 and Linux with X11 are supported by Cua Driver but not yet tested with this plugin.Agents & ProvidersBoros MátéNew
FirstMate CrewHand one agent your to-do list and it runs a crew of worker threads, each in its own git worktree, until the work lands. The first mate is a pinned thread in bb's own chat that writes each worker's instructions, supervises it, and brings back branches, pull requests, findings and only the decisions that are yours. A FirstMate tab shows the crew as a board of Queued, Working, Blocked, Parked, Done, Failed and Idle, with Steer, Interrupt, Relaunch and End on every card. A built-in watch tells the first mate within about five minutes when a pull request is merged, reviewed or changes check status. Requires bb 0.46 or later, git, and the logged-in gh CLI for the pull request watch; a capable model such as Claude Sonnet is recommended for the first mate.Agents & ProvidersGustavo AmbrozioNew
FirstmateTurn one thread into a crew: dispatch parallel agent ships in their own worktrees and land their work when it reports DONE. Ships get isolated git worktrees, scouts stay read-only, and every crew ends with a DONE, BLOCKED, or FAILED verdict. Thread events ping you the moment a crew finishes, fails, or stalls, and a crew that idles without a verdict gets nudged until it reports. Queue work with dependencies, gate merges behind green checks, and hand whole projects to domain captains. Requires the bb CLI on the host; the optional real firstmate toolbelt clones upstream firstmate scripts.Agents & Providersamrtawfik16014
GigaAM Voice InputDictate in Russian with the bb microphone and get punctuated text, recognized on your own machine. On first load the plugin installs GigaAM v3, Sber's open Russian speech model, with a private Python and CPU-only PyTorch, so nothing is set up by hand. A 10-second phrase takes about 1 second on a 10-core CPU, and recordings over 20 seconds are split at pauses, so minute-long voice notes work. Audio never leaves the machine, or you can point the plugin at your own OpenAI-compatible transcription server. Needs Linux (x86-64 or arm64) or macOS on Apple silicon, about 2 GB of disk, 1.5 GB of RAM, and a one-time 1.2 GB download; recognizes Russian only.Agents & ProvidersAiXenixNew
Jev, Please ProceedKeep agents moving when they stop only to ask for an unnecessary confirmation. Jev reviews the complete available top-level conversation whenever a thread becomes idle and continues only when its Noul value exceeds 0.95. Before sending `please proceed`, it confirms that the thread is still idle and has no pending interaction or queued message, then stops if the same final response repeats. Use a TypeSafe, Vercel AI Gateway, or OpenRouter API key; conversation text is sent through the selected provider for TypeSafe processing, and usage charges may apply.Agents & ProvidersLawrence Luk17
Kimi CodeAdds Moonshot's Kimi Code as a provider: run its threads from bb's provider picker, locally and on enrolled remote hosts. Turns stream into the native bb timeline — replies and tool calls such as file edits and shell commands — with resume and bb permission modes gating each action. The model picker fills from Kimi Code itself (K3, K3-256k, K2.8 Preview, K2.7 Code Highspeed), with thinking levels and the Default / Plan / Auto / YOLO mode selector. Kimi Code's slash commands (/compact, /status, /usage, and your own skill commands) appear in the composer menu. Requires the Kimi Code CLI and a signed-in account on each host (kimi login); the provider card can install the CLI where it is missing.Agents & ProvidersDmitry PolishukNew
LegoSplit a coding task across parallel agents and get it back assembled and verified by one Master Builder, while a Lego model builds live. The planning agent writes an instruction book: bags that each own their files, the interfaces every bag must honour, and the steps to assemble them. You choose how many builders run, up to eight, and which provider and model the builders and the Master Builder use; nothing starts until you submit. A banner above the composer and a side-panel workbench show minifigure builders placing bricks as each step lands, and finished models go on a collector's shelf. Builders and the Master Builder are ordinary BB threads on providers you already set up, so a build spends several agents' worth of tokens.Agents & ProvidersMacHatter1New
Magic ComposeChoose where a new thread runs as you type your prompt. Let Jev select the project, machine, model, reasoning level, and environment, then review or override each choice before sending. Tune its choices with your instructions and try routing a prompt from settings. Requires BB 0.43+ and a Vercel AI Gateway or OpenRouter API key; prompts, routing instructions, project context, and machine information are sent to TypeSafe through your gateway, with usage charges.Agents & ProvidersSawyer Hood38
MoA ProviderGet a second and third opinion on every message before one agent changes your files. Each preset you build becomes a model in the model picker: up to six advisor models answer in parallel, then an aggregator model reads their notes and does the work. A live panel in the thread shows what each advisor is doing and its answer as it streams. Type /moa in any other thread to put a question to the same advisors, or have the aggregator check in with them every few tool calls. Uses the providers you already installed and signed in to; each advisor adds a model turn to every message, and what an advisor reads goes to its model's provider.Agents & ProvidersMacHatter1New
Mixture of AgentsCompare two independent perspectives before your chat model acts on each request. Enable MoA per chat, share model and fallback settings across chats, and inspect each participant’s input, answer and delivery receipt from the message history. Advisors retain separate sessions, and active models can continue beyond a long-wait notice. Uses your configured BB providers and accounts, adding model calls, latency and provider usage; remote file reads require the separately installed File Gateway plugin.Agents & ProvidersKirill Vechkasov11
SoulsGive any thread a persona you define once: its job, voice, expertise and hard limits, the same in every thread that runs it. An agent interviews you in four to seven questions and puts the draft on a review card, and nothing is saved, edited or deleted until you approve. Pick a soul from the composer's + menu for an open thread or the next one, and a banner shows it with a pixel-art portrait that animates with what the agent is doing. After a context compaction the persona goes back in, and a parent thread can start a child already in character. Compare a soul with no soul on the same prompt before you rely on it. Souls stay in the plugin's own SQLite store with no account or external service, and a comparison spends model runs only when you start it.Agents & ProvidersMacHatter1New
Voice InputDictate into any BB chat and get clean text back, with your product names spelled right. Choose which machine does the recognition: Whisper (MLX) and GigaAM v3 keep the audio on that machine, while Groq, OpenAI and Google transcribe in the cloud with your own API key. A term vocabulary repairs garbled product and tool names, and an optional AI pass fixes the ones the rules miss. Recordings are written to disk before transcription, so a failed run never loses what you said. Local engines need Python 3, ffmpeg and macOS on Apple silicon; cloud engines need a paid API key.Agents & ProvidersVKirill27
Voice ModeTalk to a thread's agent hands-free, like ChatGPT or Codex voice. When the composer is empty its send button becomes a waveform; click it, or press Ctrl+Shift+V, for a full-duplex realtime voice chat. The voice model hands requests to the thread's agent, answers "is it done?", stops the agent when you say stop, and reads back a short summary when the turn ends. Talk over it any time to interrupt. Uses OpenAI Realtime with your own API key, or a plugin that provides realtime connections (for example a company AI proxy). bb's own send and dictation are never affected.Agents & ProvidersSteve FrostNew