Chat Nodes
Applies to: ChengOS v0.2.0+ ｜ Last verified: 2026-09-25 ｜ Source: crates/cheng-nodes/src/nodes/builtin/chat
chat/input and chat/output are the two ends of a conversational workflow: one receives the incoming
message and session identity, the other returns the reply on the same session. chat/history_search
is the read side — list conversations, search messages, or pull the messages around one hit — and it
is what lets an agent answer questions about what was already said without keeping the entire history
in the prompt.
Chat Input Node — chat/input
Receive multi-modal input (text, images, files, etc.), process and send to downstream nodes
Inputs
| Field | Type | Required | Default | Description |
|---|
raw_input | string (port) | ✅ | — | Raw input payload (JSON format). Connection port, not a form field. Control: textarea |
user_message_rich | object (port) | — | null | Optional rich user message from upstream nodes. Connection port, not a form field. |
context | object | — | null | Optional AgentContext restored before this workflow execution |
conversation_id | string (port) | — | null | Conversation ID (optional), generated at entry and passed through to all downstream nodes. Connection port, not a form field. |
workspace_id | string (port) | — | null | Workspace ID (optional), used for per-workspace chat isolation. Connection port, not a form field. |
document_id | string (port) | — | null | Document ID (optional), injected by document page and passed through ambient context to downstream nodes. Connection port, not a form field. |
stream_mode | boolean | — | false | Enable streaming send. Control: switch |
enable_ocr | boolean | — | false | Enable OCR for images. Control: switch |
merge_strategy | enum | — | per_file | Multi-input merge strategy. Control: select. Options: merge_all (合并所有消息为一个包), per_file (每个文件单独一个包), per_line (每行单独一个包) |
max_file_size_mb | integer | — | 50 | Maximum file size (MB). Control: number |
model_override | string | — | null | Model override (format: credential_id::model_name) |
Outputs
| Field | Type | Description |
|---|
agent_context | object (port) | AgentContext wrapper, connectable directly to agent/llm context port. Connection port, not a form field. |
user_message | string (port) | User message text. Connection port, not a form field. |
user_message_rich | object (port) | Structured user message with multimodal parts. Connection port, not a form field. |
conversation_id | string (port) | Conversation ID (passed through to downstream nodes). Connection port, not a form field. |
workspace_id | string (port) | Workspace ID (passed through for per-workspace isolation). Connection port, not a form field. |
document_id | string (port) | Document ID (passed through ambient context to downstream document nodes). Connection port, not a form field. |
stats | object | Processing statistics |
ui_event | enum | UI event |
model_override | string | Model override (format: credential_id::model_name) |
attachments_json | string (port) | Attachments JSON array, connectable directly to io/file_upload.files_json. Connection port, not a form field. |
Chat Output Node — chat/output
Receive message packets and render to chat interface, supports streaming display, virtualization and message export
Inputs
| Field | Type | Required | Default | Description |
|---|
context | object (port) | — | "" | Agent final reply text or AgentContext (auto extracts text). Connection port, not a form field. |
artifacts | object (port) | — | [] | Artifact reference(s) to render as media attachments. Connection port, not a form field. |
trace | string (port) | — | null | Agent execution trace (thinking process). Connection port, not a form field. |
theme | enum | — | system | Display theme. Control: select. Options: light, dark, system |
autoscroll | boolean | — | true | Auto scroll to latest message. Control: switch |
keep_messages | integer | — | 1000 | Maximum messages to keep in memory. Control: number |
virtualize_threshold | integer | — | 100 | Enable virtualization when message count exceeds this value. Control: slider |
display_format | enum | — | standard | Message display format. Control: select. Options: compact (紧凑模式), standard (标准模式), detailed (详细模式) |
show_timestamp | boolean | — | true | Show message timestamp. Control: switch |
Outputs
| Field | Type | Description |
|---|
reply | string | Assistant final reply text content |
trace | string | Agent execution trace (thinking process) |
ui_events | array<enum> | UI events to send to frontend |
message_count | integer | Current total messages retained |
virtualized | boolean | Whether virtualization rendering is enabled |
stats | object | Processing statistics |
Chat History Search — chat/history_search
Search past chat conversations across the workspace. Exposed as an on-demand tool to LLM/Agent — NOT a default context provider. Do not use for file paths or file contents; use file tools for those.
Inputs
| Field | Type | Required | Default | Description |
|---|
action | enum | — | search_messages | What to do: list_conversations | search_messages | get_conversation_messages | get_message_context. Options: list_conversations (列出 workspace 内活跃过的会话（按 updated_at desc）。), search_messages (按关键字 + 时间范围检索消息。), get_conversation_messages (取指定会话最近 N 条消息。), get_message_context (以某条消息为中心扩展上下文。) |
workspace_id | string | — | null | Workspace UUID. Falls back to execution metadata if absent. |
target_conversation_id | string | — | null | Restrict the search to a specific conversation. Distinct from the current session's conversation_id so cross-session searches are possible. |
query | string | — | null | Keyword (case-insensitive, ILIKE-style). |
time_range | enum | — | last_week | 时间范围。默认 last_week。 Options: last_hour, last_day, last_week, last_month, all_time |
limit | integer | — | 10 | 单次返回上限。 |
message_id | string | — | null | 锚点 message_id（仅 get_message_context 使用）。 |
context_before | integer | — | 2 | 锚点之前要带出多少条消息。 |
context_after | integer | — | 2 | 锚点之后要带出多少条消息。 |
snippet_max_chars | integer | — | 200 | snippet 最大字符数（按字符截断，避免单条消息撑爆上下文）。 |
Outputs
| Field | Type | Description |
|---|
found | boolean | Found |
no_results_found | boolean | No Results Found |
total | integer | Total |
results | array<object> | Results |
summary | string | Summary |
suggestion | string | Suggestion |
call_limit_reached | boolean | 当本次执行的搜索配额已用尽时为 true，调用方应避免再次发起。 |
Next steps