Back to Mana

Exploratory

Read as MarkdownAgent documentation indexOpenAPI JSON

Exploratory is Mana's research agent for ad-hoc questions and analyses.

What You Can Do

  • Ask financial research questions in plain English
  • Pull data from SEC filings, prices, news, earnings transcripts, company metadata, and the web
  • Get charts, tables, and reports as downloadable artifacts
  • Run multi-step research that the agent plans and tracks itself
  • Upload documents and ask questions against them
  • Ask follow-ups in the same conversation

Start a Conversation

Open the Mana app and use the chat input on the home page, or go directly to /chat/new. Type your question and send it. Follow-ups continue in the same thread, so the agent can use the context from earlier messages.

Use via the API

Create a conversation with POST /api/v1/conversations and set agentType to exploratory:

{
  "prompt": "Compare Apple and Microsoft stock performance over the past 5 years.",
  "agentType": "exploratory"
}

Useful request fields:

  • artifactIds: files uploaded first with POST /api/v1/artifacts/upload, for the agent to read.
  • responseSchema: a JSON Schema. Read the extracted result from GET /api/v1/conversations/{conversation_id}/structured-output.
  • model and reasoning: the model and reasoning effort for the run. GET /api/v1/models lists the models you can use, each with the reasoning values it accepts. Omit model to use the agent's own default. If your account is in a program with a set model, every run uses it: programModelId names it and it is the only model listed.
  • title and projectId: the conversation's title and the project it is filed in. GET /api/v1/projects lists your projects.
  • parentConversationId: send a follow-up in an existing conversation.

List your conversations with GET /api/v1/conversations. It filters by the status, pipeline_id and since query parameters, and pages with limit and cursor (pass the returned nextCursor). Stop a running turn with POST /api/v1/conversations/{conversation_id}/cancel. That stops only the conversation's own turn; POST /api/v1/conversations/{conversation_id}/stop-work also stops the agents it delegated to.

Rename a conversation with PATCH /api/v1/conversations/{conversation_id} and a title. POST .../archive hides it from your conversation list and stops what it still has running; POST .../unarchive restores it. Archive and stop-work answer 409 while a turn you just sent is still starting; once its status is running, they stop it.

See the Quickstart for full examples covering polling, streaming, questions, and follow-ups.

Example Questions

  • "Compare Apple and Microsoft stock performance over the past 5 years."
  • "Find recent 8-K filings mentioning 'restructuring' from large-cap companies."
  • "Summarize the latest earnings call for NVDA and pull out the forward-looking guidance."

Artifacts

Charts, tables, and files generated during a run are saved as artifacts. You can view artifacts inline or download them. Via the API, list them with GET /api/v1/conversations/{conversation_id}/artifacts and download each with GET /api/v1/artifacts/{artifact_id}/download.

When to Use Exploratory vs. a Pipeline

Exploratory is best for one-off questions and research threads. If you want the same analysis to run on a schedule or react to new data, build a Pipeline.