> ## Documentation Index
> Fetch the complete documentation index at: https://docs.labelbox.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Build in the console
> What each console screen is for, who can see it, and how to launch and inspect a session without writing code.
The [Recursion console](https://recursion.labelbox.com) is the visual way to work in Managed Agents: set up agents and environments, store credentials, launch sessions, and watch them work. Sessions started from the API show up here too, so you can inspect any run in one place.
## Before you begin
* Sign in with an approved account. See [Quickstart](/managed-agents/quickstart#1-sign-in).
* The console works in your tenant's **Default** organization. There's no organization or tenant switcher. See [Organizations and roles](/managed-agents/organizations-and-roles).
## The sidebar
| Screen | Use it to | Details |
| --------------------- | --------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------ |
| **Agents** | Create agents, edit their model, prompt, tools, skills, integrations, and credential access, and start sessions. | [Agent setup](/managed-agents/agents) |
| **Skills** | Create skills, group them, and publish new versions. | [Skills](/managed-agents/skills) |
| **Sessions** | Launch sessions, filter the list, open a session to follow its work, and select sessions to evaluate. | [Sessions](/managed-agents/sessions) |
| **Memory** | See what each agent learned from its sessions, its learning status and activity, and choose its memory model. | [Memory](/managed-agents/memory) |
| **Files** | Upload files to attach to sessions, and find, preview, and download the deliverables sessions saved. | [Files](/managed-agents/files) |
| **Environments** | Create sandboxes, write and test setup scripts, and set compute and internet access. | [Environments](/managed-agents/environments) |
| **Automations** | Start sessions on a schedule, from Slack, GitHub, or webhook events, or with **Run now**, and review each run. | [Automations](/managed-agents/automations) |
| **Event sources** | Create and monitor the Slack, GitHub, and custom webhook endpoints automations listen to. | [Receive events](/managed-agents/automations#receive-events) |
| **Credential vaults** | Create vaults and add tokens and secrets that agents can use. | [Vaults](/managed-agents/vaults) |
| **Integrations** | Connect GitHub, Slack, Jira, Confluence, Loom, LaunchDarkly, and catalog apps, choose tools, and check or disconnect connections. | [Integrations](/managed-agents/integrations) |
| **Analytics** | Chart usage, cost, completions, environment startup, and quality across sessions. | [Analytics](/managed-agents/analytics) |
| **Evaluations** | Create evaluators, follow evaluation runs, and compare quality by agent version. | [Evaluations](/managed-agents/evaluations) |
| **API keys** | Create and revoke your personal API keys. | [API keys](/managed-agents/api-keys) |
| **Billing** | Buy prepaid credits, manage cards, and download receipts. | [Billing](/managed-agents/billing) |
Who sees what:
* **Billing** appears only for members with the billing, admin, owner, or primary owner tenant role.
* The other screens appear for anyone whose role includes Managed Agents access. Actions that change something need the developer or admin role.
As the primary owner of your own tenant, you see every screen. A link to a screen you can't open lands on the first one you can.
Your name and email sit at the bottom of the sidebar. Click them to switch between light and dark mode, or to **Sign out**.
## Agents
Click **Create agent** to open the create form. Choose the **Blank** card to start from an empty definition, or a template card to start from a ready-made evaluator, then fill in the sections:
| Section | What it sets |
| --------------------- | --------------------------------------------------------------------------------------------------------------- |
| **General** | **Name**, **Model**, **Reasoning effort**, **Max concurrent sessions**, **Description**, and **System prompt**. |
| **Multiagent** | **Work as a team**, the subagents this agent may delegate to, and an advisor model. |
| **Skills** | Skills and skill groups the agent can load. |
| **Integrations** | The connected apps the agent's sessions may use, added with **Add integration access**. |
| **Tools** | **Web search** and MCP servers, by **Server name** and **Server URL**. |
| **Credential access** | The vaults and credentials the agent's sessions receive by default. |
| **Outcome** | A **Default rubric** that grades every session unless the launch overrides it. |
Saving an agent publishes a new immutable version. Sessions already running keep the version they started with. Open an agent and click **Start session** to launch it with that agent preselected. An agent's **Memory** tab shows what it has learned.
## Skills
Click **Create skill** to add a skill, or **New group** to create a group that attaches several skills at once. Search by name, description, or id. Archive a skill or group from its actions. See [Skills](/managed-agents/skills).
## Sessions
The list shows the sessions in the organization. Filter by **Status**, **Agent**, agent tags, or your own metadata, or search by text.
### Launch a session
Click **Launch session** on **Sessions**, or **Start session** on an agent's page to preselect it.
Under **Agent**, choose the **Agent** and the **Environment**.
Under **Credential access**, choose the vaults or individual credentials this run receives.
Under **Task**, enter the **Opening message**. To grade the result, tick **Grade this session against a rubric** and write the **Rubric**.
Under **Team**, choose whether the agent may work as a team. Under **Files**, add files from your library. Under **Custom metadata**, add up to 32 key and value pairs you can filter on later.
Click **Launch session**. The session opens while its sandbox is prepared.
The launch form changes only this session, never the agent's saved defaults. A graded launch from the console sends no `max_iterations`, so the agent revises until the grader is satisfied. To cap grading passes, start the session with the API. See [Outcomes](/managed-agents/outcomes).
### Inspect a session
The center of a session is its **Transcript**: messages, tool calls, and results as they happen. Click **Outcome** above it to see the grader's result for each pass and each criterion.
The panel beside the transcript has these tabs:
| Tab | What it shows |
| --------------- | ---------------------------------------------------------------------------------------------------------- |
| **Session** | The agent, environment, timing, usage, and outcomes. |
| **Events** | The raw event log, with the selected event's full payload. |
| **Tools** | Every tool call, with counts of completed and failed calls. |
| **Files** | Files attached to the session and the deliverables it saved, with **Attach files**, preview, and download. |
| **Evaluations** | Evaluation verdicts recorded for this session. |
| **Compute** | CPU and memory use of the sandbox over time. |
| **Work** | The session's shared task board and working notes. |
| **Threads** | Each agent in a multi-agent session. Appears only when there's more than one. |
The header shows the status, execution state, duration, and usage. From it you can:
* Click **Model cost** to see what the session and its whole tree cost.
* Click **Ask** to ask the session analyst questions about the run.
* Send a follow-up from the composer at the bottom, with **Attach image** or **Attach from Files** to include an image or document. A stopped session resumes.
* Interrupt the current turn from the live indicator while the agent works.
* Open the actions menu for **Hard stop…**, which cancels the work and tears down the sandbox, or **Delete session…**.
## Environments
Click **Create environment**, then set:
* **Name**, and **Runtime**. Choose **Agent runner** for managed compute. The runtime can't be changed later.
* **Internet access**: **No access** by default, or **Enabled**.
* **Compute**: CPU and memory, or **Let the platform size it**.
* **Storage and lifecycle**: **Workspace disk (GiB)** and **Idle stop after (seconds)**.
* A setup script, optionally started from a template, with a **Setup timeout (seconds)**.
A setup script must pass a test run before sessions can use the environment. Click **Save and test setup** to save and test it in one step. See [Environments](/managed-agents/environments).
## Memory
The table lists each agent with its **Memories** count, **Learning status**, and **Last updated** time. Open an agent to search its **Memories** by text or **Topic**, read a memory and its **History**, see each update under **Activity**, and set the **Memory model**. See [Memory](/managed-agents/memory).
## Files
Click **Upload file** to add files, with an optional **Expires** and **Purpose**. Filter by source to see **Uploads** or **Session outputs**, preview a file, download it, or open the session that produced it. See [Files](/managed-agents/files).
## Automations
Click **New automation**, choose the agent version and environment, write the **Agent instructions**, and add **Schedule**, **Slack**, **GitHub**, or **Webhook** triggers. New automations start paused: try one with **Run now**, then **Resume** it. Open an automation for its runs and **Schedule status**. See [Automations](/managed-agents/automations).
## Event sources
Create a **Slack Events API**, **GitHub webhook**, or **Custom webhook** source, choose its verification credential, and copy the **Callback URL** to the provider. The list shows each source's **Status**, **Triggers**, and **Delivery health**. Sources from connected Slack and GitHub accounts appear here too. See [Receive events](/managed-agents/automations#receive-events).
## Analytics
Choose **Usage**, **Completions**, **Environments**, or **Quality**, and a time range from **Last 24 hours** to **Last year**. Charts and the **Agents**, **Tools**, and **Models** tables cover every session in scope. See [Analytics](/managed-agents/analytics).
## Evaluations
**Overview** charts quality for one evaluator by agent version and criterion, **Evaluators** lists your evaluator agents, and **Runs** lists evaluation runs. To start one, select sessions on **Sessions** and click **Evaluate selected**. See [Evaluations](/managed-agents/evaluations).
## Credential vaults
Click **Create vault** and name it. Open the vault and click **Add credential** to store an MCP token, an environment-variable secret, or a webhook signing secret, and to limit which hosts it may reach. Secret values are write-only: the console shows how each credential is stored, never the value. See [Vaults](/managed-agents/vaults).
## Integrations
The screen lists your organization's connections under **Integration connections**, with each one's state, what uses it, and the result of its last check. Click **Add integration** to connect GitHub, Slack, Jira, Confluence, Loom, or LaunchDarkly by signing in, or to pick an app from the catalog and choose the tools agents may use. A row's menu has **Check**, **Tools**, **Reconnect**, and **Disconnect**. Then grant the app in an agent's **Integrations** section. See [Integrations](/managed-agents/integrations).
## API keys
Click **Create key**, then set **Name**, **Expires**, and **Scope**. The key is shown once. Revoke a key from its row's actions. The list shows only your own keys. See [API keys](/managed-agents/api-keys).
## Billing
The top of the screen shows your **Credit balance** and **Payment method**. Click **Buy credits** to buy prepaid credits, **Manage cards** to change cards, and follow a purchase's **Receipt** link under **Purchases**. See [Billing](/managed-agents/billing).
## What can go wrong
| Symptom | Cause | Fix |
| ------------------------------------------------------------------------ | --------------------------------------------------------------- | ----------------------------------------------------------------------------------------------- |
| **Managed Agents is unavailable**: "Your session could not be verified." | The console couldn't confirm your sign-in. | Reload the page. |
| **Billing** isn't in the sidebar | Your tenant role isn't billing, admin, owner, or primary owner. | Use an account with one of those roles. |
| **Create environment** is missing | No runtime is available to you. | Reload the page. If it's still missing, see [Troubleshooting](/managed-agents/troubleshooting). |
## Next steps
Run a graded task from start to finish.
Look up every start field and how to follow a session.
Continue, interrupt, stop, or delete a session.
Read model cost for one session or a whole tree.