Skip to main content1. Description
The Agent module in Bud AI Foundry centralizes how teams build, version, and reuse agents and prompts across projects. Agents are backed by the same GPU-optional runtime as models and deployments, so you can experiment on CPUs, burst to GPUs/HPUs when needed, and keep governance consistent across environments.
Agents ship with built-in lifecycle controls: you can search and filter the global catalog, create agents from either the Agents home or a project’s Agents tab, view “Use this agent” snippets, and manage versions from a dedicated detail page. Permissions, audit-friendly actions (Create, Edit, Delete), and deployment-aware versions keep prompts aligned with the right backends before they reach end users.
2. USPs (Unique Selling Propositions)
1. Unified catalog plus project-scoped control
Browse every agent from the global Agents page or the Agents tab inside a project, with consistent search, filters, and table actions.
2. Workflow-driven agent creation
Launch a guided drawer that collects project context, agent type, model selection, and configuration steps so teams can create production-ready agents without manual wiring.
3. Versioning with deployment context
Track current and previous versions, attach them to deployments, and edit versions directly from the detail page to keep prompts synchronized with the right endpoints.
4. Built-in reuse and integration
Use-this-agent actions expose ready-made cURL, Python, or JavaScript snippets so downstream apps and routes can consume agents immediately.
5. Governance and safety alignment
RBAC-gated actions (create, edit, delete) honor project scopes, while audit-friendly tables capture status, modality, and creation history to keep change control intact.
3. Features
3.1 Global agent catalog and discovery
- Agents home lists prompt/agent cards with model icons, descriptions, tags, and timestamps for quick scanning.
- Header search filters by name with debounce to refine results rapidly.
- Rich filters for Type (Prompt or Agent), Category, Author, Tags, and rating range help teams segment large libraries; applied filters render as removable chips.
- Grid cards include hover menus for Edit, and “See more” popovers expand longer descriptions without leaving the page.
3.2 Project-scoped Agents tab
- Inside a project detail page, the Agents tab renders a sortable table with columns for Prompt Name, Version, Type, Status, Model Name, Modality, and Created On.
- Table header offers search by name plus a Create New button (permission-gated) that jumps straight into the multi-step agent workflow for the current project.
- Row actions include Use this agent (opens integration snippets), inline Edit (loads the agent into the drawer for updates), and Delete with confirmation safeguards.
- Sorting on name, status, and created date helps teams prioritize active or recent agents.
3.3 Agent creation and editing workflow
- Create New/Agent buttons open a guided drawer that seeds a workflow ID, lets you pick project context, choose agent type, select a backing model or deployment, and configure prompts or parameters before saving.
- Structured and unstructured inputs/outputs are supported: add input and output variables (string, number, boolean, array, object) or keep the prompt freeform. Structured mode can be toggled per session to enforce schemas across runs.
- OAuth-aware flows can resume sessions after connector authorization, restoring model selections, schema variables, and workflow steps automatically.
- Edits reopen the drawer with existing content preloaded, so teams can adjust prompts, tags, or routing targets without rebuilding from scratch.
- The right-rail Tools panel lists available MCP-compatible connectors (e.g., Asana, Linear, Notion, GitHub, Atlassian) with search and categories.
- Connecting a tool opens a grant-type selector (client credentials or authorization code) so you can authorize MCP endpoints securely within the agent builder.
- Once connected, tools can be invoked in agent workflows alongside prompts and variables, enabling end-to-end orchestration without leaving the canvas.
3.5 Versions tab in agent detail
- Agent detail pages load a Versions tab showing the current default version plus a Previous Versions list.
- + Version clones context from the base prompt and opens the drawer to configure a new version without losing lineage.
- Edit on any version fetches deployment metadata when available (endpoint name mapping) so updates stay aligned with the correct backend.
- Version cards surface created/updated timestamps and endpoint names to keep operational history visible.
3.6 Integration and reuse
- “Use this agent” exposes cURL, Python, and JavaScript snippets for rapid embedding in downstream apps or routing policies.
3.7 Permissions and safety rails
- Create/Edit/Delete actions respect Bud RBAC (Project Manage) so only authorized users can change production agents.
- Delete confirmations prevent accidental removals and ensure dependent workflows can be checked before deletion.
4. How-to Guides
4.1 Accessing the Agents module
- Log in to Bud AI Foundry.
- Open Agents from the sidebar to view the global catalog of agents and prompts.
- Use the search bar to filter by name or open the filter popover to narrow by Type, Category, Author, Tags, or rating range.
4.2 Create agent
- On the Agents page, click + Agent in the header.
- The creation drawer starts a workflow for you.
- Select the target Project to set scope.
- Choose Simple Prompt (or relevant) agent type to define how the agent will run.
- In the prompt configuration step, select the Model/Endpoint the agent should call.
- Structured and unstructured inputs/outputs are supported: add input/output variables (string, number, boolean, array, object) for schema enforcement, or keep the prompt freeform for unstructured use.
- Add a system prompt and prompt messages for the required roles (system, user, and assistant), and connect MCP tools if needed for downstream actions.
- Click Run Agent to test the configuration; navigate back to prompt configuration if changes are required.
- Click Save and continue to the next step in the workflow.
- Provide Deployment Name, Tags, Description, Concurrency (minimum and maximum). Configure options like auto-scaling, caching, auto-logging, and rate limiting as needed.
- A success screen confirms deployment with model details and the URI; copy cURL/Python/JS snippets to invoke the newly deployed agent immediately.
4.3 Filter and search agents
- In the Agents page header, type a name in the search box to narrow results.
- Click the filter icon to open the filter panel.
- Set Type, Category, Author, Tags, and rating min/max; applied filters appear as chips you can remove individually.
- Click Apply to refresh the listing; Reset clears all filters.
4.4 Create an agent inside a project
- Open a project detail page and select the Agents tab.
- Click + Create New to launch the multi-step agent workflow scoped to the project.
- Complete the steps (type, model/deployment, prompt configuration) and save; the new agent appears in the table.
4.5 Use an existing agent
- Open a project detail page and select the Agents tab.
- Click Use this agent for the perferred agent.
- Copy cURL, Python, or JavaScript snippets from the drawer to integrate the agent into applications or routes.
4.6 Edit agent
- From the Agents page, use the kebab menu to edit the agent.
- Update the existing details such as name, default version number, tags and description.
- Save to update the catalog entry; changes propagate to the listing and detail view.
4.7 Delete agent
- Open a project detail page and select the Agents tab.
- Hover over the agent row and click the delete icon.
- Confirm the warning; the table refreshes after successful removal.
4.8 View and add versions
- Open a project detail page and select the Agents tab.
- Click an agent row to open its detail page.
- Review the Current Version card and Previous Version list for timestamps and endpoint alignment.
- Click + Version to clone the prompt into a new version; configure details in the drawer and save.
4.9 Edit an existing version
- Open a project detail page and select the Agents tab.
- Click an agent row to open its detail page.
- Click Edit on the desired version card.
- Preloads existing version metadata and its linked deployment.
- Update prompts or deployment references and save to keep the version synchronized.
- Default version can be modified from the Project Agents tab as well. Hover over the preferred agent and click the Edit icon.
5. FAQ
Q1. Where can I manage agents within a project?
Use the Agents tab on the project detail page to search, sort, create, edit, delete, and “Use this agent” without leaving the project context.
Q2. How do filters work on the Agents page?
Filters support Type, Category, Author, Tags, and rating ranges; selections appear as removable chips, and search combines with filters for precise discovery.
Q3. Can I resume agent creation after OAuth or connector steps?
Yes. The workflow restores sessions (including selected models and schema variables) after OAuth callbacks so you can continue configuration without losing state.
Q4. How do I keep prompts aligned with deployments?
Edit versions from the Versions tab; Bud fetches deployment details when available so you can confirm endpoint names before saving.
Q5. What permissions are required to modify agents?
Create, Edit, and Delete actions require manage-level permissions (Project Manage) per Bud RBAC. Users without permissions can still search and view agents where allowed.
Q6. Can I work with multiple agent versions at once?
Yes. The agent builder supports multiple simultaneously open versions/sessions (up to the concurrent session limit) so you can edit, compare, and update side-by-side before saving.
Q7. Do agents support structured schemas and MCP tools?
Agents can enforce structured input/output schemas (with variable types and validation) or stay unstructured. MCP-compatible tools can be connected through the Tools panel and authorized via client credentials or authorization code to run inside workflows. 