Manage independent GovTribe AI conversations
Start, inspect, continue, wait briefly for, stop, delete, and organize independent GovTribe AI workstreams, including their model and reasoning settings.
GovTribe AI can create and manage separate conversations when work should continue as an independent workstream. These conversation-management tools are internal GovTribe AI function tools, not public GovTribe MCP tools exposed to connected clients.
Choose the right kind of work
| Need | Use |
|---|---|
| Continue analysis, drafting, or follow-up in the current chat. | Keep working in the current conversation. |
| Start a separate or parallel workstream that a user can open and continue independently. | Create another AI conversation. |
| Run work later, on a schedule, or after a supported event. | Create an automation. |
Conversation labels such as manager, worker, researcher, or reviewer are instructions in the message. They do not create a parent-child relationship or a persisted coordination graph.
Start and organize independent work
Use Read_AI_Conversation_Project without a project_id to find existing projects before creating a duplicate. Create a project only when it materially helps group related conversations.
Call Create_AI_Conversation once for each independent workstream. Give its initial message a self-contained objective, useful context, constraints, expected output, and any role label that helps the conversation work independently. The call can also set a name, description, project, model, and reasoning level.
Record the returned conversation ID so later calls can inspect or update the same workstream. Do not invent manager, worker, parent, child, join, or work-item IDs.
Use the conversation and project tools
| Tool | What it does |
|---|---|
Create_AI_Conversation | Creates and starts one independent conversation with an initial message and optional name, description, project, model, and reasoning level. |
Read_AI_Conversation | Returns one conversation's metadata, work state, project, generation settings, credit usage, and permitted user/assistant messages. |
Update_AI_Conversation | Changes metadata, project assignment, model, or reasoning level; sends a message; or stops active work. |
Delete_AI_Conversation | Stops active work when needed, then removes the conversation from normal use. |
Create_AI_Conversation_Project | Creates one owned conversation project. |
Read_AI_Conversation_Project | Lists owned projects or reads one project with paginated conversation summaries. |
Update_AI_Conversation_Project | Renames one owned project. |
Delete_AI_Conversation_Project | Deletes one project and unassigns its conversations without deleting them. |
Wait_For_AI_Conversations | Waits up to ten minutes inside the current response for one to eight owned conversations to finish or need attention, then returns bounded status, links, and newly visible messages after a cursor. |
These tools can manage only conversations and projects owned by the current user in the current workspace. Hidden internal conversations are excluded.
Choose a model and reasoning level
Create and update calls accept these model values:
gpt-5.6-solgpt-5.6-terragpt-5.6-luna
They accept none, low, medium, or high for reasoning_level.
For Create_AI_Conversation, an omitted or null model or reasoning level inherits the user's current GovTribe AI default. For Update_AI_Conversation, an omitted or null value preserves the conversation's current selection. Response length is not changed by these tools.
An update made while a conversation is working applies to its next response. The returned generation_settings.active_turn continues to describe the response already in progress, while model, reasoning_level, and generation_settings.selected show the conversation selection.
Inspect status and messages
Use Read_AI_Conversation to check response_status, is_working, is_streaming, project assignment, model, reasoning level, generation settings, and credit usage. A project read also includes model and reasoning level in each conversation summary.
The read tool returns the latest 20 visible messages by default and accepts a limit from 1 to 100. Use before_message_id to request an older page. Messages are user and assistant conversation content only; tool-call outputs, hidden setup messages, and internal memory output are excluded.
Reading another conversation's messages requires Full Conversation Recall. Without it, the tool still returns metadata and work state, with messages unavailable.
Continue, steer, or stop work
Send message with Update_AI_Conversation to start an idle conversation or steer an active one. Do not combine message and stop in the same call. A conversation cannot message, stop, or delete itself while its current tool call is running.
After a create, message, or stop mutation, inspect the returned operation.type and operation.result. Treat fields such as started, steerAccepted, queued, or requested as the authoritative receipt. Report any returned reason or billing blocker instead of claiming the requested action succeeded.
Check status again only when a fresh update serves the user's immediate request, or use the bounded wait described below when the result belongs in the current response. These tools do not provide a durable join, automatic continuation, exact-once resume, or a queue that wakes an inactive conversation when another conversation finishes later.
Wait briefly for other conversations
Call Wait_For_AI_Conversations when the current response should include another conversation's result. Give it one to eight owned conversation IDs as targets, an optional after_message_id for each target, a return_when of any_terminal or all_terminal, and a timeout_seconds from 0 for an immediate snapshot up to ten minutes. The wait ends when the condition is met, a target needs attention, the timeout expires, or the user sends a new message to or stops the current conversation.
Each target returns a wait_state of pending_start, working, completed, stopped, incomplete, failed, or unavailable, plus needs_attention and an attention_reason of billing_blocked, failed, recovery_failed_retryable, recovery_failed_terminal, stalled, unanswered_user_message, deleted, or not_found, the conversation url, and, when Full Conversation Recall permits, new_messages after the cursor in transcript order, ten at a time. Pass the returned next_after_message_id back as after_message_id on the next call so a repeated wait does not report unchanged output as new, and keep calling while has_more is true. When cursor_matched is false the cursor was not found and the page is the newest visible output rather than new output. Targets reported as unavailable never satisfy the wait. ended_by reports condition_met, timeout, snapshot, source_interrupted, or capacity_unavailable.
The wait is foreground coordination inside the current response. It persists nothing, creates no parent-child link or callback, and never resumes an inactive conversation later. Use an automation when monitoring must outlive the current response.
Rename, move, and delete safely
Use Update_AI_Conversation to rename a conversation, change its description, or assign it to a project. Supplying a null description clears it, and supplying a null project_id unassigns the conversation.
Use Update_AI_Conversation_Project to rename a project. Deleting a project unassigns its conversations but does not delete them. Deleting an active conversation stops its work first.
Delete a conversation or project only when the user explicitly requests that exact deletion. Report the returned ID and deletion receipt after the call completes.
Related articles
- Create a conversation: Start and configure a GovTribe AI conversation in the product.
- Manage conversations: Rename, organize, reopen, and remove conversations in GovTribe.
- Tune GovTribe AI: Manage default model, reasoning, response length, and conversation recall preferences.
- Automations: Use scheduled or event-triggered AI work instead of an independent conversation.
Favorites, saved searches, and pursuits
Choose between favoriting a record, saving a search, and creating a pursuit when an agent needs to keep something for later.
MCP servers
Choose the right GovTribe MCP server before loading tool definitions, especially when an agent can route among standard, compact, compatibility, and grouped GovTribe endpoints.