Build and Run a Roundtable

Institution Mode lives under Institution in the Web UI at /institution. Its main workspace is the Roundtable, where a question moves through a configurable sequence of specialist Agents before one final report is produced.

A Roundtable performs research only. It does not place orders or move funds. Any action based on its conclusion must begin in a separate request.

Start with the default Roundtable

1. Open Institution from the sidebar.
2. Create or open an Institution session.
3. Enter a ticker, a basket, or a natural-language investment question.
4. Send the message to run the default seven-stage Roundtable.

Examples:

Analyze BTC for a 6–12 month holding period.

Compare BTC, ETH, and SOL. Which has the strongest risk-adjusted setup?

Review NVDA before earnings. Focus on expectations, valuation, and downside risk.

If Minara cannot identify the asset confidently, it asks you to choose the intended ticker before spending the full research budget.

Open the Roundtable Builder

Select Roundtable in the Institution header. The Builder has three working areas:

• Library on the left for Roundtable, phase, and Agent templates.
• Roundtable preview in the center for the phase order and participants.
• Settings on the right for the selected phase or Agent.

The header shows a budget estimate and validation issues. Undo and Redo apply to the current editing session.

Compose the phases

Select a phase card to edit its title, execution type, shared prompt, and Debate rounds. You can:

• insert a new phase before Final;
• drag phases or use the arrow controls to reorder them;
• duplicate or delete a non-Final phase;
• load a built-in or saved phase template;
• save the current phase as a reusable template.

Choose the phase type according to how work should flow:

The Final phase is locked to the end. A Debate needs at least two Agents. Synthesis and Final each need exactly one manager, and Final contains exactly one Agent.

Add and customize Agents

Select an Agent card to open its settings. You can change:

1. Name and role title to make responsibility clear in the live Roundtable.
2. Framework to state the investment method or analytical lens.
3. System prompt to preserve the template, customize it, or start from a blank instruction set.
4. Skills to expose all eligible research skills or a selected subset.
5. Web research to allow or prevent web search for that Agent.
6. Loop behavior to use one pass or several iterations, with explicit limits for tool calls, timeout, and output length.

Use Compiled prompt to inspect the effective instruction before running. It combines the Agent instructions, the phase prompt, the task context, and the required output contract.

Agents can be added, duplicated, reordered within a phase, or removed when the phase minimum remains satisfied. Applying an Agent template replaces the selected Agent's framework, prompt, skills, and loop settings. Save a useful configuration as an Agent template when you want to reuse it elsewhere.

A practical stock research Roundtable

To decide whether a public company deserves a long-term investment, you might build:

The point is not to maximize the number of Agents. Each Agent should own a distinct question that could change the decision.

Save the right level

• Select Save session to make the edited Roundtable active for the current Institution session.
• Select Save template to preserve the whole Roundtable for later sessions.
• Save an individual phase or Agent when only that component should be reusable.
• Use Restore to return the working copy to its source template.

Built-in templates cannot be changed. User-created Roundtable templates can be renamed or deleted, and user-created phase templates can be deleted. Existing sessions and run snapshots retain their own copies, so deleting a template does not erase past work.

Run and inspect the work

After saving, close the Builder and submit the research question. The Roundtable sidebar shows:

• the active phase and Agent;
• pending, running, complete, failed, or skipped status;
• live reasoning, tool collection, review, and writing activity;
• each Agent's conclusion and detailed research report;
• chronological run history for the session.

The final response is a synthesis of the available evidence, not a transcript of every Agent. Open an Agent result when you need to inspect its sources, tool activity, assumptions, or disagreement with the final conclusion. HTML and PDF report files appear with the completed run when rendering succeeds.

Every run freezes the Roundtable used for that decision. If you edit the session afterward, the next run uses the new configuration while the old run keeps its original phases, Agents, limits, and outputs.

Single asset and baskets

For one asset, the session Roundtable runs once and the Final Agent produces the report. For a basket, Minara runs the per-asset research and adds an Allocator that compares the usable results or proposes weights. The basket result can be partial when one asset fails, and the failure remains visible.

Compare BTC, ETH, and SOL for the next quarter.

Allocate $50,000 across BTC, GLD, NVDA, and USDC with a moderate risk budget.

Read the result carefully

The default template aims to produce a rating, thesis, scenarios, trade plan, and risk review. A custom Roundtable follows the prompts and roles you configure, so its final format may differ. Check:

• whether the run is complete, partial, or insufficient_data;
• which phases or Agents failed;
• what evidence was actually collected;
• what assumptions would reverse the conclusion;
• whether the proposed action fits your own portfolio and risk limits.

Acting on a conclusion

Institution Mode never moves money. Begin execution with a separate prompt, for example:

Use the BTC base case from the Roundtable and prepare a $1,000 spot order.

Minara resolves the order again, shows a preview, and waits for the required confirmation. A Roundtable recommendation is evidence for your decision, not permission to trade.

CLI and operations

The Web UI is the full Roundtable customization surface. CLI commands remain useful for running the default Institution workflow and managing the learning lifecycle:

minara institution history --limit 20
minara institution history --ticker BTC
minara institution finalize <run_id> --reason "position closed"
minara institution reflect <run_id> --window 7d
minara institution reflect-pending

Environment limits and learning controls are documented under Institution Mode environment variables.

Next steps

• Institution Mode concepts and guarantees
• Learning system
• Operator reference