Create, configure, and manage Projects in Studio. Projects are the top-level container for agents, tools, knowledge bases, workflows, and deployment configuration on Agent Platform.
Studio
Studio is the primary web interface for building, testing, deploying, and monitoring AI agents on the Agent Platform. It provides a unified workspace where teams can manage every phase of the agent lifecycle, from initial design through production operations.
When you sign in, Studio displays the Projects dashboard - a grid of project cards showing all projects in your workspace. Each card displays the project name, description, orchestration pattern badge, agent count, and last-updated timestamp.
From the dashboard, you can:
- Search projects using the search bar at the top of the page.
- Create a new project using the “New Project” option.
- Open a project by clicking any project card, which navigates into that project’s workspace.
Projects
A project is the top-level container in Studio. It groups agents, supervisors, tools, knowledge bases, workflows, integrations, and configuration into a single workspace, and it defines the isolation boundary for deployments, access control, and analytics.
What a Project Contains
Every agent you build belongs to exactly one project. A project holds seven categories of resources and their associated settings.
| Category | Contents |
|---|
| Build | Agents, workflows, and the project overview. |
| Resources | HTTP tools, code sandbox tools, MCP servers, knowledge bases, integrations. |
| Evaluate | Evaluations and experiments. |
| Operate | Sessions, deployments, inbox, alerts, transfer sessions. |
| Insights | Dashboards, analytics, agent performance, quality monitoring, customer insights, feedback, voice analytics, and pipelines. |
| Govern | Guardrails and governance policies. |
| Settings | Members, API keys, models, config variables, Git, runtime config, and more. |
Before You Begin
Before creating or managing projects, confirm the following.
- Workspace access. You must be signed in to Studio with a user account that belongs to the target workspace.
- Role. Creating projects, managing members, and deleting projects require administrator permissions on the workspace. Developers without admin rights can still build inside a project they have been added to.
- LLM provider (optional). To use AI-guided project creation, a workspace administrator must configure an LLM provider under Admin > Models. If the provider is not configured, the AI Architect panel shows an in-panel banner instead of the wizard.
Accessing the Projects Dashboard
When you sign in, Studio opens the Projects dashboard. This is the entry point for all project management operations.
The dashboard displays a grid of project cards. Each card shows the project name, description, orchestration pattern badge, agent count, and last-updated timestamp.
From the dashboard you can:
- Search existing projects using the search bar in the header.
- Create a new project from the New Project dropdown.
- Open a project by clicking its card, which loads the project workspace.
- Access the three-dot menu on any card to delete that project.
To return to the dashboard from anywhere in Studio, click the product logo in the top-left corner or follow the breadcrumb trail back to Projects.
Creating a Project
Studio supports three creation methods. Choose the one that fits your starting point.
| Method | When to Use It |
|---|
| AI-guided (Architect) | You are new to the platform, or you want the AI Architect to scaffold agents from a described use case. |
| Blank project | You know exactly which orchestration pattern and agents you need, and you want full manual control. |
| From template | A predefined template matches your use case (for example, customer service or IT help desk). |
Method 1: AI-Guided Creation
Use this method when you want the AI Architect to scaffold agents, tools, and project structure from a described use case. It is the recommended starting point for new users.
To create a project with the AI Architect:
- On the Projects dashboard, click New Project > Start with AI Architect.
- Read the Welcome screen and click Next to begin the interview.
- In the Interview phase, answer the Architect’s questions about your use case, target audience, channels, and success criteria.
- In the Upload phase, optionally attach documents that describe your domain — for example, knowledge base content, API specifications, or existing agent scripts. The Architect uses these to inform agent and tool design.
- Wait for the Generating phase to complete. The Architect produces agent definitions, tool configurations, and an initial project structure.
- In the Reveal phase, inspect the generated structure and adjust agent roles, names, or connections as needed.
- In the Review phase, confirm that the configuration is correct.
- Click Create. Studio provisions the project and opens its workspace.
The richer the context you provide during the interview and upload phases, the more accurate the generated agents will be. Upload real documentation rather than summaries where possible.
Method 2: Blank Project
Use this method when you want manual control over every aspect of the project.
To create a blank project:
-
On the Projects dashboard, click New Project > Blank Project.
-
In the creation dialog, enter the following fields.
| Field | Required | Description |
|---|
| Name | Yes | A descriptive project name. Appears on the dashboard card and in breadcrumbs. |
| Description | No | A short summary of the project’s purpose. |
| Orchestration pattern | Yes | The coordination model for agents in this project. See the table below. |
-
Select an orchestration pattern based on your use case. Refer to Choosing an orchestration pattern.
-
Click Create. Studio opens the project workspace with an empty overview.
Method 3: From Template
Use this method to start from a predefined project that already contains agents and resources for a common use case.
To create a project from a template:
- On the Projects dashboard, click New Project > From Template.
- Browse the template catalog and select a template that matches your use case.
- Review the included agents, tools, and orchestration pattern.
- Provide a name and optional description.
- Click Create. Studio clones the template into your workspace as a new project.
Templates come pre-configured with the orchestration pattern and agent roles appropriate for the use case, so you typically only need to edit prompts, tool credentials, and knowledge sources before testing.
Choosing an Orchestration Pattern
The orchestration pattern determines how agents coordinate with each other. It is set at project creation time and pre-configures agent roles and coordination blocks for new agents.
| Pattern | How It Works | Best For |
|---|
| Concierge (recommended) | A single front-facing agent handles the user and delegates to specialist agents behind the scenes. | Customer service, help desk, general-purpose assistants. |
| Router | A triage agent classifies the request and transfers the user directly to the appropriate specialist. | Multi-department support, intent-based routing. |
| Tiered | A triage agent routes to L1 agents, which can escalate to L2 agents. Supports de-escalation back to a lower tier. | Technical support with defined escalation paths. |
| Custom | No predefined structure. Agents can be connected in any topology. | Advanced use cases requiring flexible orchestration. |
Choosing a pattern does not lock you in. It only pre-configures coordination defaults. You can still add, remove, or reconnect agents within a project at any time.
Navigating a Project Workspace
Clicking a project card opens the project workspace. The workspace has three persistent elements: the header bar, the project sidebar, and the main content area.
Project Overview Page
The first page you see inside a project is the Overview. It adapts to the maturity of your project.
| Project State | What You See on the Overview |
|---|
| Empty (no agents) | A get-started prompt guiding you to create your first agent or import existing agents. |
| Building (agents, no deployments) | Metric cards for agent count, session count, and deploy ratio; an agent list; and a resource summary panel. |
| Live (active deployments) | Full metrics for sessions, messages, tokens, cost, and deployments; active deployment list; activity timeline; and quick action cards. |
| Item | Purpose |
|---|
| Overview | Adaptive project home page with metrics, agent list, and quick actions. |
| Agents | Agent list, agent editor, and in-editor testing. |
| Workflows | Workflow builder and execution monitoring. |
| Item | Purpose |
|---|
| Tools | HTTP tools, code sandbox tools, and MCP servers. |
| Knowledge Bases | Document ingestion, indexing, and search configuration. |
| Integrations | Third-party connector catalog and active connections. |
| Group | Pages |
|---|
| Evaluate | Evaluations, Experiments. |
| Operate | Sessions, Deployments, Inbox, Alerts, Transfer Sessions. |
| Insights | Dashboard, Analytics, Billing and Usage, Agent Performance, Quality Monitor, Customer Insights, Feedback, Voice Analytics, Agent Transfer, Pipelines. |
| Govern | Guardrails, Governance. |
| Settings | Members, API Keys, Models, and more. |
Project Switcher
The project switcher dropdown sits at the top of the sidebar and lets you move between projects without returning to the dashboard. It shows a scrollable list of all your projects with the current project highlighted.
Command Palette
Press Cmd + K (macOS) or Ctrl + K (Windows/Linux) to open the command palette. The palette provides fuzzy search across:
- Navigation — jump to any page within the current project.
- Agents — open a specific agent by name.
- Actions — reset session, clear events, and other quick commands.
The command palette is the fastest way to navigate large projects. Type a few characters of a page or agent name to filter results instantly.
Monitoring Project Health
The project Overview aggregates the key runtime metrics for the project. Use it as your first stop when checking project health.
| Metric | What It Measures |
|---|
| Agent Count | Total number of agents in the project. |
| Session Count | Conversations processed by the project’s agents. |
| Messages | Total messages exchanged across all sessions. |
| Tokens | LLM token usage across all agents. |
| Estimated Cost | Cost computed from token consumption and configured model pricing. |
| Deployed Ratio | Number of agents with active deployments versus total agents. |
Exporting and Importing Projects
Export and Import let you move agent definitions and configuration between projects or environments. Both actions are available from the project Overview.
Export Project
Use the Export Project dialog to package an agent project into a portable .zip file. The archive contains the project’s agents, tools, and any optional layers you select—such as guardrails, workflows, and channels. The archive can be imported into another environment or saved as a versioned snapshot.
Before exporting, review the project contents in the export dialog.
- Project Summary: Overview of the project, including project name and number of agents and tools in the project.
- Agents: List of agents in the project.
- Tools: List of tools in the project.
- Export Warnings: If the platform detects any conditions that affect the export, such as automatic normalization of tool formats, it surfaces them in the Export Warnings. All warnings are informational and do not block the export process. Review warnings before exporting to understand what the platform has automatically adjusted.
- Export Layers: Control which parts of the project are included in the output archive.
Click Export to download a .zip archive to your browser. The file follows the standard folder structure(shown above) and can be imported directly using the Import Project dialog.
- Credentials must always be reconfigured in the target environment. The export only includes the structural requirements and metadata (such as which profile types/auth fields are needed).
- Any hardcoded secret values remaining in text/YAML files are replaced with
{{config.X}} template references before archiving.
Import Project
Use the Import Project dialog to import an agent project or individual agent files into the Agent Platform. The import supports merge-based import, stale content removal, environment reference adaptation, and three-way conflict detection.
Navigate to Project > Overview > Import.
Upload Import File
Drag a file into the upload area, or click the upload area to open a file browser. The platform accepts the following formats:
| Field / Option | Description | Notes |
|---|
.zip | A compressed archive containing a full project folder structure. | Most common for full project imports. |
.agent-bundle.json | A JSON bundle that packages agent definitions and metadata. | Produced by the platform’s Export function. |
.ablenc | An encrypted ABL (Agent Behavior Language) file. | Used for secure distribution of agent logic. |
.abl | A plain-text ABL source file for a single agent. | Individual agent import. |
.yaml / .yml | A YAML-formatted agent or configuration file. | Individual agent or config import. |
.zip archive. The contents of the zip are mapped according to the subfolders.
your-project/
├── project.json # Manifest
├── abl.lock # Integrity
├── agents/ # Core - agent definition
│ ├── supervisor.agent.abl
│ └── booking_agent.agent.abl
├── tools/ # Core - tool definition
│ └── hotels-api.tools.abl
├── config/ # Core
├── connections/ # Connections
├── guardrails/ # Guardrails
├── workflows/ # Workflows
└── locales/ # Optional
└── en/
└── booking_agent.json
Import Mode
The Import Mode section controls how the incoming project is reconciled with the target project in the platform. The default mode is always Merge.
Merge with existing project (default)
Matching records from the archive are updated in the target project. Records that exist in the target but are absent from the archive are left untouched. This is a non-destructive operation; no existing content is deleted unless explicitly overridden by the Replace modifier below.
Replace project contents
When checked, this modifier changes how each import layer (e.g., agents, tools, connections) is handled: any content belonging to that layer that is present in the target but absent from the archive is removed. Content that exists in both the archive and the target is updated normally.
Replace does not switch the import to a full overwrite. It is a modifier on top of Merge. Records that are in both the archive and the target are merged; records only in the target are deleted per layer; records only in the archive are added.
| Record exists in | Merge only | Merge + Replace |
|---|
| Archive only | Added to target | Added to target |
| Both archive and target | Updated (merged) | Updated (merged) |
| Target only | Kept unchanged | Removed from target |
Adapt References for This Environment
When enabled, the platform performs post-import normalization steps to make the imported project compatible with the current tenant and environment.
- Portable bindings - Resolves any environment-portable connection or resource references so they point to the correct endpoints for this tenant.
- Eval references - Normalizes evaluation dataset or metric references so they are valid within the current environment’s evaluation framework. If a referenced modelId doesn’t exist in the target tenant, the user is prompted to pick a replacement.
- Post-import validation - Runs a validation pass on the imported project after the above steps complete, surfacing any remaining reference errors like auth profiles, environment variables, connector credentials, MCP server auth, etc.
Use a Base Export for Conflict Checks
Enables three-way conflict detection during import. When active, the platform compares three states.
- The base snapshot - the original export of the project before any edits were made.
- The archive being imported - the version with the author’s changes.
- The current target - the live state of the project in the platform.
When enabled, attach the base export file. This allows the platform to distinguish between a true conflict (where both the archive and the target have independently changed the same record from the base) versus a non-conflicting divergence (where only one side changed). Without a base snapshot, any difference between archive and target is treated as a conflict.
How to Import a Project
- Navigate to the project where you want to import content.
- Click Import.
- Drag your file into the upload area, or browse and select a file. Accepted formats:
.zip, .agent-bundle.json, .ablenc, .abl, .yaml.
- Choose your Import Mode options. Merge is always active. Check Replace project contents if you want to remove layer-level content from the target that is absent from the archive.
- Optionally, check Adapt references for this environment to normalize bindings and references for the current tenant.
- Optionally, check Use a base export for conflict checks to enable three-way conflict detection.
- Click Import to start the import.
- Review any validation errors or conflict reports surfaced by the platform after import completes.
Deleting A Project
Deleting a project removes it and all of its contents permanently.
To delete a project:
- Open the Projects dashboard.
- Click the three-dot menu on the card of the project you want to delete.
- Select Delete.
- In the confirmation dialog, type the project name exactly as shown.
- Click Delete.
Deleting a project permanently removes all agents, tools, knowledge bases, workflows, sessions, and deployment configurations in that project. This action cannot be undone. Export the project first if you need a backup.
Features and Permissions
Some project features depend on workspace configuration or your assigned role. If a feature appears unavailable, check the conditions below.
| Feature | Requirement |
|---|
| AI Architect | An LLM provider must be configured under Admin > Models. Otherwise an in-panel banner is shown. |
| Platform Admin link | Visible only to users with super-admin permissions. |
| Admin pages (Members, Models, Security) | Require workspace administrator permissions. |
| Some Insights pages | Agent Performance, Quality Monitor, and Customer Insights may show a “coming soon” placeholder if not yet enabled for your workspace. |
