> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.band.ai/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.band.ai/_mcp/server. # Adapter Features > Reference for band-sdk 6.0.1 adapter construction, the shared config and constructor options, the emit, capabilities, and tool filter keywords, and the SUPPORTED_EMIT and SUPPORTED_CAPABILITIES of every adapter. Every built-in adapter is built from one typed config object plus keyword-only live objects, and every one accepts the same five feature keywords. This page owns those shared rules; the per-adapter pages list only adapter-specific settings. ## Adapter Construction ```python XAdapter( config: XAdapterConfig | None = None, *, history_converter=..., # adapters that take one additional_tools=..., # adapters that take them ..., # live objects: clients, graphs, factories, callbacks **features: Unpack[FeatureKwargs], ) ``` `config` holds plain settings: model names, prompts, timeouts, and modes. Live objects stay keyword-only constructor arguments, so a config can always be loaded from YAML or JSON with `XAdapterConfig.model_validate(data)`. `config` is required only when no usable default exists: `A2AAdapter` needs `remote_url`, `ACPClientAdapter` needs `command`, and `SlackAdapter` needs `apps`. Every `XAdapterConfig` is a frozen pydantic model. An unknown field raises `ValidationError` (`Extra inputs are not permitted`), so a removed field name fails at construction instead of being ignored. > **Warning** > > Adapter constructors take no flat settings. `AnthropicAdapter(model="claude-sonnet-5-5")` raises `TypeError: SimpleAdapter._resolve_features() got an unexpected keyword argument 'model'`. Pass `AnthropicAdapter(AnthropicAdapterConfig(model="claude-sonnet-5-5"))`. ## Common Adapter Options The prompt options `custom_section` and `system_prompt` are fields on the adapter's config object. `history_converter` and `additional_tools` are constructor arguments, and the `**features` keywords are constructor keywords. The `**features` keywords are accepted by A2A, A2A gateway, ACPClient, Agno, Anthropic, ClaudeSDK, Codex, CopilotACP, CopilotSDK, CrewAI, CrewAI Flow, CursorACP, Gemini, GoogleADK, LangGraph, Letta, NOOA, OmpACP, OpenCode, Parlant, PydanticAI, Slack, and Strands. `BandACPServerAdapter` accepts the keywords but declares no emit kinds or capabilities, so a nonempty `emit` or `capabilities` raises `BandConfigError`. | Option | Applies to | Description | | :----------------------------- | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :---------------------------------------------------------------------------------------------------- | | `custom_section` (config) | ACPClient, Anthropic, ClaudeSDK, Codex, CopilotACP, CopilotSDK, CrewAI, CursorACP, Gemini, GoogleADK, LangGraph, Letta, NOOA, OmpACP, OpenCode, Parlant, PydanticAI, Strands | Additional instructions added to the adapter prompt | | `system_prompt` (config) | Anthropic, Codex, Gemini, GoogleADK, Parlant, PydanticAI, Strands | Full prompt override where supported | | `emit` | Any adapter taking `**features` | Narrow which event kinds the adapter reports into the room; see [Adapter Features](#adapter-features) | | `capabilities` | Any adapter taking `**features` | Add platform tool groups to the schemas the model sees; see [Adapter Features](#adapter-features) | | `include_tools` | Any adapter taking `**features` | Keep only these platform tools by name | | `exclude_tools` | Any adapter taking `**features` | Drop these platform tools by name | | `include_categories` | Any adapter taking `**features` | Keep only platform tools in these categories: `chat`, `contacts`, `memory`, `files` | | `history_converter` (argument) | Agno, Anthropic, ClaudeSDK, Codex, CopilotSDK, CrewAI, CrewAI Flow, Gemini, GoogleADK, LangGraph, Letta, NOOA, OpenCode, Parlant, PydanticAI, Strands | Convert Band room history into the framework-specific format | | `additional_tools` (argument) | ACPClient, Anthropic, ClaudeSDK, Codex, CopilotACP, CopilotSDK, CrewAI, CrewAI Flow, CursorACP, Gemini, GoogleADK, LangGraph, NOOA, OmpACP, OpenCode, Parlant, PydanticAI, Strands | Add framework-compatible custom tools | `additional_tools` takes `CustomToolDef` pairs, `(InputModel, handler)`, where `InputModel` is a pydantic model describing the arguments; LangGraph also accepts LangChain tools, and Pydantic AI and Strands also accept plain callables. A custom tool declares what a successful call does to the turn with `declares_turn_effect`; see [Custom Tools and Turn Effects](/reference/sdk/platform-tools#custom-tools-and-turn-effects). ## Adapter Features Adapters take their Band feature settings as direct keyword arguments, typed as `**features: Unpack[FeatureKwargs]`. There are five keys, all optional: | Key | Type | Default | Effect | | :------------------- | :----------------------------------- | :------------------------------ | :---------------------------------------------------------------------------------- | | `emit` | `Emit \| Iterable[Emit]` | Everything the adapter supports | Event kinds reported into the room timeline | | `capabilities` | `Capability \| Iterable[Capability]` | None | Platform tool groups added to the schemas the model sees | | `include_tools` | `Iterable[str]` | All | Keep only platform tools with these names | | `exclude_tools` | `Iterable[str]` | None dropped | Drop platform tools with these names | | `include_categories` | `Iterable[str]` | All | Keep only platform tools in these categories: `chat`, `contacts`, `memory`, `files` | ```python from band import Capability, Emit from band.adapters import AnthropicAdapter, AnthropicAdapterConfig config = AnthropicAdapterConfig(model="claude-sonnet-5-5") # Report only token usage, and add the contact tools adapter = AnthropicAdapter( config, emit={Emit.USAGE}, capabilities={Capability.CONTACTS}, ) # Emit is flag-capable, so a union, a set, or a list all work adapter = AnthropicAdapter(config, emit=Emit.TOOL_CALLS | Emit.USAGE) # Silence the adapter completely adapter = AnthropicAdapter(config, emit=()) ``` `Emit` and `Capability` members: | Value | Effect | | :-------------------- | :------------------------------------------------------------------------------------------------------------------ | | `Emit.TOOL_CALLS` | Emit `tool_call` and `tool_result` events into the room timeline | | `Emit.TASK_EVENTS` | Emit task lifecycle events | | `Emit.THOUGHTS` | Emit the agent's intermediate reasoning as `thought` events | | `Emit.USAGE` | Emit turn token-usage events | | `Capability.CONTACTS` | Add the contact tools | | `Capability.MEMORY` | Add the memory tools | | `Capability.TASKS` | Add the seven task board tools | | `Capability.FILES` | Gates the room file tools, which this reference does not cover; see [Platform Tools](/reference/sdk/platform-tools) | Combine members of one enum with `|`; mixing the two raises `TypeError: cannot combine Emit with Capability`. [Platform Tools](/reference/sdk/platform-tools) lists the tools in each group and what each group needs on the platform side. > **Note** > > `emit` and `capabilities` default differently. Omitting `emit` resolves to every kind the adapter declares in `SUPPORTED_EMIT`, so events are on unless you narrow them; an explicit empty value such as `emit=()` silences the adapter. Omitting `capabilities` adds nothing, because each capability puts extra tool schemas in front of the model on every turn. The `capabilities` argument on the [schema accessors](/reference/sdk/platform-tools#agenttoolsprotocol) is a separate layer with its own default. > **Warning** > > Passing an `Emit` or `Capability` member the adapter does not declare raises `BandConfigError` at construction, naming the unsupported values and the supported set, for example `A2AAdapter does not support emit kind(s): thoughts; supported: (none)`. The adapter is not built. At startup, `Agent.start()` drops any capability the connected deployment does not serve. As of `band-sdk` 6.0.1 this applies only to `Capability.FILES`, which is dropped unless the agent's `/me` response reports the `ff_file_transfer` feature flag as on. The three tool filters narrow the platform tool schemas, in strict precedence: `include_categories`, then `include_tools`, then `exclude_tools`. Each stage narrows the result of the previous one, so `include_categories=["chat"]` together with `include_tools=["band_store_memory"]` yields nothing, because `band_store_memory` is in the `memory` category. Unknown names in `include_tools` or `exclude_tools` are logged as a warning (`include_tools contains unknown names: ...`) and otherwise ignored. > **Warning** > > `Emit.TASK_EVENTS` carries state on the Codex, Letta, and OpenCode adapters: each persists its session, thread, or agent resume mapping in task-event metadata gated by that flag. It is in the default `emit` set for all three, so leaving `emit` alone is safe. An explicit `emit=` replaces that default, so any set you pass must still include `Emit.TASK_EVENTS`, and `emit=()` stops resumption across restarts. ### What Each Adapter Supports `SUPPORTED_EMIT` is also the default `emit`, so this table doubles as what the adapter reports when you pass no `emit`. An adapter with an empty emit set emits nothing and rejects every nonempty `emit` value. | Adapter | `SUPPORTED_EMIT` | `SUPPORTED_CAPABILITIES` | | :--------------------- | :-------------------------------------- | :------------------------------------- | | `A2AAdapter` | none | none | | `A2AGatewayAdapter` | none | none | | `BandACPServerAdapter` | none | none | | `ParlantAdapter` | none | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `CrewAIAdapter` | `TOOL_CALLS` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `CrewAIFlowAdapter` | `TOOL_CALLS` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `AnthropicAdapter` | `TOOL_CALLS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `GeminiAdapter` | `TOOL_CALLS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `GoogleADKAdapter` | `TOOL_CALLS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `LangGraphAdapter` | `TOOL_CALLS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `PydanticAIAdapter` | `TOOL_CALLS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `StrandsAdapter` | `TOOL_CALLS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `ACPClientAdapter` | `TOOL_CALLS`, `THOUGHTS`, `TASK_EVENTS` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `CopilotACPAdapter` | `TOOL_CALLS`, `THOUGHTS`, `TASK_EVENTS` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `CursorACPAdapter` | `TOOL_CALLS`, `THOUGHTS`, `TASK_EVENTS` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `OmpACPAdapter` | `TOOL_CALLS`, `THOUGHTS`, `TASK_EVENTS` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `AgnoAdapter` | `TOOL_CALLS`, `THOUGHTS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `ClaudeSDKAdapter` | `TOOL_CALLS`, `THOUGHTS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `CopilotSDKAdapter` | `TOOL_CALLS`, `THOUGHTS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `NooaAdapter` | `TOOL_CALLS`, `THOUGHTS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `LettaAdapter` | `TOOL_CALLS`, `TASK_EVENTS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `OpencodeAdapter` | `TOOL_CALLS`, `TASK_EVENTS`, `USAGE` | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `CodexAdapter` | all four | `CONTACTS`, `MEMORY`, `TASKS`, `FILES` | | `SlackAdapter` | the inner adapter's | the inner adapter's | `BandACPServerAdapter` declares no emit kinds or capabilities, so there is nothing to narrow. On the ACP client adapters, `emit` controls which session updates reach the room: `TOOL_CALLS` posts `tool_call` and `tool_result` events (including the pair that records an approval request), `THOUGHTS` posts agent thoughts, and `TASK_EVENTS` posts plan updates. Agent text is not gated. The `task` event that records the room's session mapping is posted whatever `emit` says, so `session/load` resume survives `emit=()`. `SlackAdapter` declares no sets of its own: it adopts and validates against the inner adapter's features, so read its row off the adapter you wrap. `ParlantAdapter` accepts `Capability.MEMORY` but offers no memory tools, so that capability adds nothing. `NooaAdapter` posts the model's raw `reasoning_content` as thoughts, so pass `emit` without `Emit.THOUGHTS` to keep reasoning out of the room. ## Adapter Summary | Adapter | Purpose | Required or primary inputs | Reference | | :----------------------------------- | :---------------------------------------------------------------------------------------------- | :------------------------------------------------------------------------------------------------------------------ | :----------------------------------------------------------------------- | | `LangGraphAdapter` | LangGraph agents, as a ReAct agent built from a chat model or your own graph | `llm`, with an optional `checkpointer`, or `graph_factory` / `graph` | [Frameworks](/reference/sdk/adapters/frameworks#langgraphadapter) | | `PydanticAIAdapter` | Pydantic AI agents with type-safe tools | `PydanticAIAdapterConfig(model=...)` or `llm` | [Frameworks](/reference/sdk/adapters/frameworks#pydanticaiadapter) | | `CrewAIAdapter` | CrewAI agents with role, goal, and backstory | Optional `CrewAIAdapterConfig` (`role`, `goal`, `backstory`) | [Frameworks](/reference/sdk/adapters/frameworks#crewaiadapter) | | `CrewAIFlowAdapter` | One CrewAI Flow run per inbound room message, with orchestration state rebuilt from task events | `flow_factory` | [Frameworks](/reference/sdk/adapters/frameworks#crewaiflowadapter) | | `GoogleADKAdapter` | Google Agent Development Kit agents | Optional `GoogleADKAdapterConfig` (`model`) | [Frameworks](/reference/sdk/adapters/frameworks#googleadkadapter) | | `AgnoAdapter` | An existing Agno agent | `agent` | [Frameworks](/reference/sdk/adapters/frameworks#agnoadapter) | | `StrandsAdapter` | Strands Agents with native tool support | `StrandsAdapterConfig(model=...)` or `llm` | [Frameworks](/reference/sdk/adapters/frameworks#strandsadapter) | | `NooaAdapter` | NVIDIA Object Oriented Agents, with Band tools as methods on `self.room` | `agent`, a `NooaRoomAgent` subclass | [Frameworks](/reference/sdk/adapters/frameworks#nooaadapter) | | `LettaAdapter` | Letta agents with persistent memory | Optional `LettaAdapterConfig` | [Frameworks](/reference/sdk/adapters/frameworks#lettaadapter) | | `ParlantAdapter` | Parlant behavioral engine | Optional `ParlantAdapterConfig` (`name`, `description`) and `nlp_service`; or your own `server` and `parlant_agent` | [Frameworks](/reference/sdk/adapters/frameworks#parlantadapter) | | `AnthropicAdapter` | Anthropic SDK with a manual tool loop | Optional `AnthropicAdapterConfig` (`model`, `provider_key`) | [Model SDKs](/reference/sdk/adapters/model-sdks#anthropicadapter) | | `GeminiAdapter` | Google Gen AI SDK with a manual tool loop | Optional `GeminiAdapterConfig` (`model`, `provider_key`) | [Model SDKs](/reference/sdk/adapters/model-sdks#geminiadapter) | | `ClaudeSDKAdapter` | Claude Agent SDK (Claude Code) with MCP tools and room approvals, one workspace per room | Optional `ClaudeSDKAdapterConfig` (`model`, `permission_mode`, `cwd`) | [Coding Agents](/reference/sdk/adapters/coding-agents#claudesdkadapter) | | `CodexAdapter` | OpenAI Codex app-server over JSON-RPC, one process per room | Optional `CodexAdapterConfig` | [Coding Agents](/reference/sdk/adapters/coding-agents#codexadapter) | | `OpencodeAdapter` | OpenCode server sessions, one per room | Optional `OpencodeAdapterConfig` | [Coding Agents](/reference/sdk/adapters/coding-agents#opencodeadapter) | | `CopilotSDKAdapter` | GitHub Copilot SDK sessions, one per room | Optional `CopilotSDKAdapterConfig` | [Coding Agents](/reference/sdk/adapters/coding-agents#copilotsdkadapter) | | `ACPClientAdapter` | Any external ACP agent process, one per room | `ACPClientAdapterConfig(command=...)` | [Coding Agents](/reference/sdk/adapters/coding-agents#acpclientadapter) | | `CopilotACPAdapter` | GitHub Copilot CLI over ACP | Optional `CopilotACPAdapterConfig` | [Coding Agents](/reference/sdk/adapters/coding-agents#acp-presets) | | `CursorACPAdapter` | Cursor's `agent acp` server over ACP | Optional `CursorACPAdapterConfig` | [Coding Agents](/reference/sdk/adapters/coding-agents#acp-presets) | | `OmpACPAdapter` | OMP's `omp acp` server over ACP | Optional `OmpACPAdapterConfig` | [Coding Agents](/reference/sdk/adapters/coding-agents#acp-presets) | | `A2AAdapter` | A remote A2A agent as a Band participant | `A2AAdapterConfig(remote_url=...)` | [Protocols](/reference/sdk/adapters/protocols#a2aadapter) | | `A2AGatewayAdapter` | Band peers exposed as A2A HTTP endpoints | Optional `A2AGatewayAdapterConfig` (`gateway_url`, `port`) | [Protocols](/reference/sdk/adapters/protocols#a2agatewayadapter) | | `BandACPServerAdapter` / `ACPServer` | Editor-facing ACP server | None; credentials come from the running `Agent` | [Protocols](/reference/sdk/adapters/protocols#acp-server) | | `SlackAdapter` | A Band agent bridged into Slack threads | `SlackAdapterConfig(apps=...)` and `inner` | [Protocols](/reference/sdk/adapters/protocols#slackadapter) |