Design, prove, and ship AI agents — in one workspace.

The agent lifecycle#

Every Studio project follows the same path. The sidebar groups its tabs into Design, Test, and Ship (Build sits under Ship), so you always know what comes next.

Your first agent#

Studio is where you design an AI agent, test it, and run it as a service. An agent is an AI assistant with a job description: instructions, a model, and the tools it may use. A project holds your agents, the workflows that chain them, their connections and their tests. Algen Agent Runtime is the engine underneath that runs them; you rarely deal with it directly.

1. Create a projectOn the projects page, open azAI and describe the agent you want, or choose New project and fill in the form.
2. Connect a modelNew projects start on Studio's built-in test model, which gives fixed answers. Add a provider such as OpenAI on the Connections tab and store its key under Credentials.
3. Try itChat with your agent in the Playground.
4. Prove itAdd test cases and run them on the Evaluations tab.
5. Ship itDeploy it as a service your apps can call, or publish it.

Sign in & organizations#

Hosted Studio uses your existing identity provider. There are no Studio-specific passwords to create or rotate.

Sign-in page (sample)

• 1Sign-in providers — use the Google or GitHub account your organization invited.

Signing in#

1. Open your Studio addressYour administrator shares the Studio URL for your organization. Unauthenticated visits land on the sign-in page.
2. Choose a providerSelect Continue with Google or Continue with GitHub. Use the account your organization invited.
3. Return to StudioAfter the provider confirms your identity, Studio opens where you left off — or the invitation you were accepting.

Accepting an invitation#

Organization admins invite you by email. The invitation link opens a preview showing who invited you and to which organization.

1. Open the invitation linkIf you are not signed in, Studio asks you to sign in first and then returns you to the invitation.
2. AcceptYou join the organization with the role chosen by the person who invited you.
3. Confirm if your email differsIf you signed in with a different email address than the one invited, Studio shows both addresses (partially masked) and asks you to confirm before joining.

Switching organizations#

If you belong to more than one organization, use the organization switcher at the top of the sidebar ( organization name). Projects, members, credits, and permissions are all scoped to the active organization.

Tour of Studio#

Studio has two persistent regions: a sidebar for navigation and a header for the project you are editing.

Studio layout (schematic)

• 1Organization switcher — change the active organization.
• 2Project switcher — jump between projects, create, or import.
• 3Workspace header — breadcrumb, search, project health, save state, azAI, and the project menu.
• 4Runtime status — shows the installed Algen Agent Runtime version. Hover for the Studio version too, or click to open the capability inspector.
• 5Account menu — your name and role. It opens Organization settings, Billing and usage, Session and security, the User guide, Runtime documentation, and Sign out. Platform administrators also see Platform console.

Sidebar sections#

Some tabs appear only when they apply. Code shows when the project has code files. Data shows when your Studio server offers data sources. A tab your role cannot use stays visible but disabled; hover it to see why.

Workspace header#

• Breadcrumb — select Projects to go back to all projects, or the project name to open its Overview.
• Search — Search or jump to… opens the command palette (⌘K).
• Project health — a chip reads Valid, or shows how many issues or warnings there are. Select it to open the Overview and see them.
• Build state — when managed builds are on, a pill shows ready building build failed or saved, build required.
• Save state — with no edits it reads Saved · rev N. After you change something it reads Unsaved changes and shows Discard and Save (⌘S).
• azAI — the sparkle button opens azAI.
• Project menu (⋯) — Validate configuration, Export as ZIP, Fork project, Rename, Archive, and Delete permanently. Items your role cannot use are greyed out.

Command palette & focus mode#

Press ⌘K (CtrlK on Windows and Linux) to search every section, your projects, and actions — save, discard, validate, export, fork, create or import a project, switch theme, or inspect runtime capabilities. Use the arrow keys and ↵ to pick a result. Press ⌘. to enter focus mode, which hides the sidebar and header so you can concentrate on the editor; press it again or Esc to exit.

Use the theme control in the sidebar footer to switch between light, dark, and system themes.

Studio and Runtime versions#

The Versions section at the top of the capability inspector lists the Studio version, the installed Algen Agent Runtime version, and the Runtime range this Studio supports. The values come from the server, so they describe the Studio you are connected to. Include them when you report a problem.

azAI, your assistant#

azAI is the assistant built into Studio. Open it with the gold azAI button in the header, or press ⌘I (CtrlI on Windows and Linux). It docks on the right of the page you are on.

azAI in Agent mode (sample)

• 1azAI button — opens and closes the panel; ⌘I does the same.
• 2Mode — Ask explains and suggests; Agent makes the changes.
• 3Steps — what azAI read or did; expand them to see each one.
• 4Approval — anything that costs or changes configuration waits for you.

What azAI does#

Approvals and undo#

azAI asks before anything that costs time or credits or changes your project's configuration, and shows what it is about to do. Choose Allow, Always allow here for that kind of action in the conversation, or Decline. Deploying, publishing, pushing and deleting history ask every time. Before a bigger build, such as a new project with several agents and a workflow, azAI shows its plan first: approve it once and azAI builds all of it, carrying on inside the new project, without asking again for each change; Checks, builds and runs still ask. You can attach a text file, such as a specification or a configuration file, with the paperclip. Each file or setting it changes can be kept or undone. azAI never sets secret values, never approves custom workflow code, and never deletes or archives a project; it tells you where to do those yourself.

Things to ask#

• “I'm new — how does Studio work?” or “What does the Data tab do?”
• “Create a support agent for my online shop that uses gpt-4o-mini.”
• “Why can't I run my agent yet?”
• “Add three test cases for refund questions and run them.”
• “Run my workflow with a sample question and tell me what it said.”

Conversations#

Conversations are private to you and kept per project; they continue where you left off. The menu at the top of the azAI panel lists your earlier conversations. Pick one to reopen it, choose New conversation to start fresh, or delete one you no longer need.

In hosted Studio, azAI's model calls use your organization's AI credits (see Credits & billing). The panel footer shows what the conversation has cost so far. Set Limit there to cap the credits one conversation may use.

Projects#

A project holds everything for one agent product: its agents, workflows, connections, evaluation cases, builds, and deployments. Projects belong to your organization, so every member with access sees the same project.

All projects (sample)

• 1New project — create a project from a form or a template. Use Import for an example, a folder, or a GitHub repository.
• 2Active / Archived — switch between current and archived projects.
• 3Card menu — rename, archive, restore, or delete a project.

Create a project#

On All projects (G H), select New project. Each card shows its agent count and Ready or Needs attention.

If your organization has no projects yet, the page instead shows three starting points:

The Create a new project dialog works like this:

1. Name the projectEnter a Project name. Studio derives a Project ID (lowercase, hyphenated) that you can edit before creating; it cannot be changed later.
2. Describe itOptionally add What should it do? and Starting instructions for the agent. You can change everything later.
3. Pick a starting pointUnder Start with, choose A single agent or Agent + starter workflow. Or pick one of the templates under Or start from a template with custom code (see below).
4. CreateSelect Create project. The project opens on its Overview.

Rather describe what you want? Select Let azAI set it up at the top of the dialog, and azAI builds the project from your description.

Templates with custom code

These templates include Python code, so you work on them in the Code tab and run Checks.

Import a project#

Select Import on the projects page and choose a source:

• This computer — upload a folder containing project.json, agent.yaml, or runtime_settings.yaml. Imports are limited to 5 MiB.
• GitHub — import from an HTTPS GitHub repository URL. You can choose a branch, tag, or commit and a subdirectory.
• Example — start from a bundled sample.

Private repositories: in hosted Studio, an organization admin first connects GitHub under Settings › Integrations and grants access to the repository. Then import it by URL like any other repository. Local Studio uses your own Git credentials instead.

On import, Studio classifies the project as declarative (configuration only) or custom code (includes Python code). Custom-code projects must be built before they can run in hosted Studio; declarative projects, including ones you create in Studio, run without a manual build.

Manage projects#

Overview & launch checklist#

The Overview is the home page of each project. A progress ring shows how many launch steps are done (for example 4 of 6 steps done), and a Recommended next step card takes you straight to the most useful action.

The Launch checklist marks each step as done, to do, or needing attention. Each open step has a button that takes you to where you do it. The steps are:

1. Describe what your agent does
2. Connect a language model
3. Add API keys and secrets
4. Resolve configuration issues
5. Prove it works with test cases
6. Ship it

The At a glance tiles count your Agents, Workflows, Model connections, Tools enabled, Test cases, and Missing keys. Select a tile to open the matching tab.

Project overview (sample)

• 1Recommended next step — one click to the most useful action.
• 2Launch checklist — done, to do, or needs attention; each step links to where you do it.
• 3At a glance — counts of agents, workflows, model connections, tools, test cases, and missing keys.

Configuration issues#

Studio checks your project each time you save. Any problems appear in the Configuration issues card. For each issue:

• Open takes you to the tab where you fix it.
• Fix appears when Studio can repair it for you. Save or discard your changes first.
• Fix with azAI asks azAI to repair it.

Studio validates every tool, planner, verifier, context builder, guardrail and model provider an agent names against what the project actually registers. Tools and components your own code registers come from a plugin the project declares in agent.yaml; Studio learns them when the project is built, so until then they wait under Build the project to check these. A name that nothing declares is an error; its Fix uses the closest registered name, never by removing a guardrail.

Code and Studio#

For a project with code, the Code and Studio card says what that code does that Studio does not run, found by reading it, never running it: its own server or workers, tools passed to build_container instead of a plugin, agents defined in code, and environment variables read directly instead of through env://. Fix with azAI moves such tools into a plugin, or such reads to env:// references.

The side of the page also has How Studio works, with links to each tab, and two shortcuts: Try it opens the Playground, and Deploy opens Deploy.

Agents#

An agent is a model plus instructions, tools, and limits. A project can contain several agents, each specialized for one job.

Agent editor (sample)

• 1Agent list — each agent with its description and model. Select one to edit it.
• 2Form / YAML — two views of the same definition.
• 3Provider and model — pick from the providers set up in Connections.
• 4Try in playground — chat with the saved agent.

1. Select or add an agentPick an agent in the list, or select Add agent.
2. Write system instructionsExplain the role, tone, boundaries, and expected output in plain language. Be specific about what the agent must never do.
3. Choose the modelPick a Provider. The agent uses that provider's default model unless you turn on Override provider model. Providers are set up in Connections.
4. Check its toolsThe Tools section lists the tools the agent may use, what each one changes (reads, writes, sends or deletes), whether it asks for approval first under the agent's safety settings, and what provides it. To change them, select Manage tools; tools and knowledge are chosen in Connections›Tools & knowledge.
5. Save and try itSave, then select Try in playground.

In the Form view, the bar at the top jumps to each part of the agent:

Switch between the Form view and the YAML view at any time. Both edit the same definition; YAML is useful for bulk changes and code review.

Workflows#

Workflows chain agents and steps into a pipeline. The canvas shows each step and its dependencies so you can see how work flows from input to final answer.

Workflow run paused at an approval step (sample)

• 1Input & run — a form appears instead when the workflow defines inputs.
• 2Canvas — live status of every step; click one to inspect it.
• 3Checkpoint panel — when a run pauses for a person, the decision controls appear beside the canvas.

Build a workflow#

The bar above the canvas picks the workflow. Its buttons let you New workflow, Rename workflow, Duplicate workflow, or Delete workflow. A workflow that another workflow uses cannot be deleted. Inputs and output sets the fields a run asks for; with no inputs, a run takes free text.

Select Add step to add a step. Agent step and Custom code step come first; More step types… shows the rest.

Select a step on the canvas to edit it: its Step name, what it Runs after, and when to Run only if. Approval steps also ask What should the reviewer check? and what to do If rejected. Save the project to keep your changes.

Run a workflow#

1. Pick the workflowUse the workflow selector if the project has more than one.
2. Provide inputWorkflows with inputs show a form; otherwise enter free text.
3. Start the runSelect Run workflow (or press ⌘↵). Studio checks admission (credits, concurrency, rate limits) and queues the run on managed infrastructure.
4. Follow progressThe run moves through queued → running → completed, failed, or cancelled. Select Cancel run to stop it.

Inspect each step#

Click any step on the canvas to open its detail drawer: the agent that ran, status, model calls, tokens, cost, and latency, followed by input, output, tool calls, memory, artifacts, events, and approvals.

Run history#

Run history lists past runs, newest first. Select a run to see it on the canvas and open its details. You can:

• Filter the list by status (All statuses, Running, Completed, Failed, Cancelled).
• Select two runs to compare them: Candidate versus baseline shows the configuration, revision, latency, tokens and cost side by side.
• Select Copy link to share a link to the selected run.

Approvals & clarifications#

A Human approval checkpoint pauses the workflow until someone decides. The panel beside the canvas shows what the step proposes:

• Approve — continue as proposed.
• Approve changes — edit the Reviewed parameters (JSON), then continue. This appears only when the step lets the reviewer edit.
• Reject — stop this path.

You can add an optional Decision note, which is recorded with the decision. When a step needs more information instead, a Human input checkpoint asks a question. Type Your answer and select Answer and continue.

Approve custom workflow code#

If a workflow uses the project's own Python code, that code must be approved before runs load it. The Workflows tab then shows Custom workflow code needs review. The approval applies to one revision, so changing the code needs another review.

In local Studio you approve it yourself: enter a reason of at least 10 characters and select Approve for each module. In hosted Studio an admin of your organization approves it. Once approved, the notice reads Custom workflow code approved.

Code & Checks#

Most of a project is configuration you edit on the Agents, Workflows, Data and Connections tabs. The Code tab is for the parts that need Python: custom workflow steps, hooks that prepare a request or check an answer, and code tools. Your files live in src/, their tests in tests/, and packages in the project's dependencies.

Code tab and a Check (sample)

• 1Project files — your Python in src/ and its tests in tests/.
• 2Editor — edits stay in your draft until you save a version.
• 3Check — saves, locks packages, builds and runs your tests.
• 4Stages — each stage's result; a failing test can be handed to azAI to fix.

Add code to a project#

The Code tab appears once a project has code. To add code to a project that has none, open the Workflows tab and select Add custom logic. Or start from a template with custom code.

Work in the editor#

The Code tab works like a small code editor. The bar on the left switches between Explorer (your files), Source control (when available), Packages, and Checks.

• Edits save as you type, but builds and runs use the last saved version. Select Save version to make your edits part of the project. Revert to saved throws away edits since then.
• Ask azAI starts a question about the open file, or the lines you selected in it. To have azAI change the code and run Checks until it works, switch the azAI panel to Agent.
• The status bar shows whether your version is saved.

Packages#

List the Python packages your code imports. Select Add package, enter its name and, optionally, a version range, then Save packages. Select Lock packages to pin exact versions so every build installs the same ones.

Git#

In Source control, select Connect a repository and enter the Repository address, plus an optional Branch or tag and Project folder. Then:

• Pull from Git brings in the code from the repository.
• Push to Git sends your saved version, with a Branch and Commit message. Save your code as a version first. When a pull request is opened, a link to it appears.

Private GitHub repositories need the GitHub connection under Integrations.

Checks#

A Check proves your code works. Select Check at the top of the Code tab. It runs these stages in order and shows each result:

1. ValidateThe project and its code are well formed.
2. Lock packagesDependencies are resolved and locked.
3. Save versionThe draft is saved as a new version.
4. Build imageAn image is built from that version.
5. Run testsYour tests in tests/ run against it.
6. ApprovalCustom code that workflows call must be approved before it runs. You review and approve it on the Workflows tab (see Approve custom workflow code); in hosted Studio an admin of your organization approves it.

If tests fail, the Checks view lists them with links to the failing lines. Select Fix with azAI to have azAI fix them.

Projects without code don't need Checks: configuration is validated when you save it.

Run tests#

Run tests runs your tests in seconds, without a build. They run in the project's last image, which already has its packages installed, with the code and tests in your editor right now, including changes you have not saved as a version.

• Run tests at the top of the Checks view runs every test in tests/.
• With a test file open, Run file runs that file and Run test runs the test your cursor is in.
• Results appear as a tree, one file per row. Select a test to open it; a failure links to the line that failed. Run one test or file again from its row.
• Fix with azAI asks azAI to fix the failing tests. Run them again when it is done.

Tests run isolated, as in a Check: no network, no project secrets, read-only code, and limits on CPU, memory and time (10 minutes). Your organization can have 2 test runs going at once and 100 a day. Runs use no credits.

A build is needed first when the project has no image yet, or when its packages changed since the last image. The Checks view then says so; select Build with Check, then run the tests. Run tests appears only for projects with code and test files under tests/.

Data#

The Data tab lists the databases your agents may query and the rules every query follows. Studio always connects read-only, keeps connection strings encrypted, and applies the query rules (which tables and columns are allowed, row limits and timeouts) to every query. Use a database user that can only read.

Data tab (sample)

• 1Add database — name it and give its connection string as an env:// reference.
• 2Database — always read-only; the connection string is stored encrypted.
• 3Query rules — which tables agents may read and how much a query may return.

To let an agent answer from documents instead of a database, add knowledge (retrieval) on the Connections tab.

Connections#

Connections hold everything your agents plug into. They are grouped into five areas:

Connections — Models & API keys (sample)

• 1Areas — the five groups of connection settings.
• 2Test provider connectivity — the refresh button confirms Studio can reach the provider.
• 3Credentials — store the value behind each env:// reference.

Add a model provider#

1. Choose a providerUnder Add another provider, pick from the catalog — for example OpenAI, Anthropic, Azure OpenAI, Mistral, DeepSeek, Hugging Face Inference, or any OpenAI-compatible endpoint.
2. Set the default modelAnd, where applicable, the base URL.
3. Reference the keyAPI keys are stored as secure references such as env://OPENAI_API_KEY, never as plain text in the project.
4. Save and store the credentialEnter the value under Credentials (below) and select Store credentials.
5. Test itSelect the refresh button on the provider (Test provider connectivity) to confirm Studio can reach it. Save your changes first.

The built-in test model#

A new project starts on Studio's built-in test model (provider mock, model deterministic). It needs no key and costs nothing, but it gives fixed answers instead of real ones, so it is only for trying Studio out. To get real answers, add a provider such as OpenAI or Anthropic above, choose its model on each agent (or ask azAI to switch it), and store its key under Credentials.

Connectors#

Connectors are ready-made tools from Algen Agent Tools: Gmail, PostgreSQL, SMTP email and more. When your Studio offers them, they appear under Connections›Tools & knowledge›Connectors, as cards with each connector's maturity (alpha connectors carry a badge), its tools and what each one changes.

• Add a connector to the agents you choose. The preview shows the tools, the permissions they need and the sites the connector calls, which are added to the project's allowed HTTP hosts. azAI can add one for you and asks first.
• Fill in its settings right after. Secrets are stored like any credential and never shown again. A database connector's server must be declared under Organization›Database hosts before hosted runs reach it.
• Build when Studio asks: until a build describes the connector, its tools show Pending build. Some Studios run connectors on their own image, with no build.
• Remove takes the connector's tools, the permissions only they needed and its sites out of every agent. Saved settings stay.
• Review upgrade appears when the project's connectors were added at an older Tools version; it lists removed tools, new permissions and new settings before anything changes.

SMTP email cannot run on hosted Studio yet. Run it from a local Studio or a deployment.

Credentials#

Your API keys and passwords are kept safe: the project, its files, exports and deployments only ever contain a reference such as env://OPENAI_API_KEY, never the value, and azAI never reads values.

The Credentials panel lists every secret reference the project uses, which components use it, and whether a value is configured. Values are encrypted with AES-256-GCM in Studio's secure vault and are never shown again after you store them — you can only replace them.

• Runtime credentials ready — everything is configured.
• 2 credentials required — enter the missing values before running.

Credentials are reused automatically by the Playground, evaluations, and deployments — configure them once per project.

Playground#

The Playground is a chat interface for one agent at a time. Each reply shows its status, and you can expand the activity to see model calls, tokens, and cost.

Playground (sample)

• 1Missing keys — appears only when a key the agent needs is not stored yet.
• 2Response — status, answer, and run details such as model calls and tokens.
• 3Composer — disabled while you have unsaved changes.

1. Save your changesThe Playground always runs the saved project. With unsaved edits it shows a notice with Save now, and sending is disabled.
2. Pick an agentUse the agent selector.
3. Check keysIf a key is missing, a notice names it, for example Add OPENAI_API_KEY in Connections before running this agent. Select Open connections to add it.
4. Send a messageType or select a suggestion, then select Send or press ⌘↵. Select Stop to end a run early.
5. Review the answerWatch the Activity panel while it runs, then copy the response or inspect the details.

Select New chat to start a fresh conversation. Recent runs lists earlier runs; select Show older runs to see more. Nothing you do in the Playground changes your project.

Attach files#

An agent can read images and PDFs you attach, such as a photo of an ID document or a scanned invoice. The model sees the file itself, so pick a model that reads images (and PDFs, for PDF files).

1. Let the agent take filesOn the Agents tab, open the agent, and under Safety & behavior turn on Accept attached files. Choose how many files a run may take, then save.
2. Attach and sendIn the Playground, select the paperclip next to the message box, pick PNG, JPEG, WebP, GIF or PDF files, and send your message. Each file shows as a chip; select its × to remove it.

For a workflow, open the step whose agent should read the files, and under Advanced turn on Give this step the run's files. The paperclip then appears above Run workflow.

Studio checks each file's real type from its contents, not its name, and refuses files a model cannot read, such as HEIC photos or images larger than 5 MB or 8,000 pixels a side. A run takes at most 10 MB of files. Uploads expire after a day.

When the agent pauses#

• Approval needed before a tool runs — the agent wants to use a tool that needs a person's approval. Select Approve, Reject, or open Adjust tool parameters, edit them, and select Approve with changes.
• The agent needs more information — type your answer and select Answer and continue.

Only roles that can approve runs can approve or reject (see Permission matrix).

Evaluations#

Evaluations turn the conversations you tried by hand into repeatable checks. Run the suite after every meaningful change to catch regressions before your users do.

Evaluations (sample)

• 1Latest result — pass rate and change since the previous run.
• 2Run N tests — runs every case against the saved project.
• 3Case checks — what must and must not appear in the answer.

Write test cases#

Select Add test case to add one case. To add many at once, select Import and choose a .csv, .tsv or .jsonl file (up to 5 MB). Pick the Question column and, optionally, the Expected answer column, then select Import test cases.

Run the suite#

Save first: tests run against your saved agents. Then select Run N tests (for example Run 6 tests). The summary shows how many passed, the pass rate, and the change since the previous run (▲ or ▼). Each case shows Passed or Failed and which check failed.

Run history lists earlier runs with their pass rate, and notes whether the project changed since the previous run. Open Compare with an earlier run to see which cases are now passing, now failing, still passing, or still failing.

Build#

Build packages a project's custom code into a signed, scanned container image. The source, dependency lock, scan results, SBOM, provenance, and signature all stay bound to that image.

Build (sample)

• 1Readiness — source, dependencies, and selected image.
• 2Actions — only the steps the project still needs are shown.
• 3Build history — select a ready build for the current revision; expand for evidence.

Build an image#

For a configuration-only project, the Build tab shows two cards: Configuration (no custom code to build) and Platform image, which Studio prepares when you save or run a valid configuration. There is nothing for you to do.

For a project with custom code, three readiness cards show where you are: Source, Dependencies, and Selected image. You can also lock packages from the Code tab.

1. Resolve dependenciesIf the project needs a dependency lock, select Resolve dependencies. Follow any Changes required before building guidance.
2. Accept the lockWhen resolution is ready, review it and select Accept lock.
3. BuildSelect Build project image. The build is scanned and signed in an isolated builder.
4. Select the buildWhen the build is ready, select Select for revision N so runs and deployments use it.

Expand any entry in Build history to see its evidence: SBOM and attestation digests, findings by severity, and recent logs. You can Cancel an active build or Rebuild a finished one.

Saving changes creates a new project revision. If the project contains custom code, build and select again before running the new revision.

When only the code changed since the last ready image (same dependency lock and platform image), the build reuses that image's installed packages and replaces only the project code, so it takes seconds rather than minutes. The history shows it as Code-only image · Reused packages from build …. Every check still runs on the new image: scan, SBOM, provenance and signature. If the earlier image was removed or differs, Studio builds in full automatically.

In the Playground and on the Workflows tab, a revision that needs a new image only because its code or Studio changed offers Build and run: one click builds the image, shows its progress, and starts the run once the image is ready. If the build fails, you see why, with Fix with azAI. Nothing is built until you click.

Publish#

Publish submits your agent to the Algen Agent Hub marketplace. Packages are quarantined and inspected before they become discoverable; no code runs during review.

Publish (sample)

• 1Marketplace link — set up once and reused for every release.
• 2Release details — five sections to complete.
• 3Submit for review — sends the package to quarantine and review.

1. Link your marketplace onceEnter the Marketplace API URL and a Marketplace API token created from the Marketplace creator dashboard, then select Link marketplace. Studio encrypts the token and reuses it; it is never added to the project or exported package.
2. Confirm the checklistThe agent passes validation, all secrets are env:// references, you have read the marketplace review standards, and you have rights to all included code and models.
3. Complete the releaseFill in the sections: Listing, Ownership & contract, Delivery, Provenance & links, and AgentStripes result. Under Delivery, choose the Delivery type: Algen package, Source container, or Hosted agent.
4. SubmitSelect Submit for review. Studio packages the release and sends it for review.

To use a different marketplace account, select Change link. Select Publish history to see earlier releases of this listing.

AgentStripes check#

Some marketplaces require an AgentStripes result before they list an agent. AgentStripes reads your agent's code on GitHub and rates its capabilities.

1. Push your codeAgentStripes reads the code on GitHub, so push your latest commit first.
2. Run the checkIn AgentStripes result, enter the GitHub repository (and an Agent folder only when the repository holds more than one agent). Select Check with AgentStripes. It takes about 30 seconds.
3. Attach itThe signed result is attached to the release. You can also choose Paste a result instead, and copy a README badge with Copy Markdown.

What Studio adds to your repository

Each push also writes the files AgentStripes reads, from your project's settings: a README.md (what the agent does and for whom, its inputs and outputs, limits, what success looks like, and how to run it), a LICENSE, SECURITY.md, CHANGELOG.md, agentstripes.yaml and, when the project has test cases, evals/cases.yaml. Studio rewrites them on every push so they stay current. To keep your own version of one, add it in the Code tab or delete its Written by Algen Agent Studio line; Studio then leaves it alone. Until you choose a licence, LICENSE says all rights reserved.

Getting ready with azAI

Ask azAI to get this agent ready for AgentStripes. It checks the project against each capability AgentStripes rates, at no cost, and proposes the changes for you to approve: clearer agent instructions, the description and limits the README needs, test cases, and tracing. It asks you only for the licence and a security contact. Then push and run the check.

Deploy#

Deploy turns a project into a running service your apps can call. You pick where it runs, fill in only the fields that target needs, review the plan, and confirm. Nothing is created until you confirm.

Deploy (sample)

• 1Keys — deployments reuse the keys stored in Connections. The banner says if any are missing.
• 2Target — where the service runs.
• 3Review and deploy — preview the plan, tick the confirmation, then deploy.
• 4Progress — status and log output; cancel while it is in progress.

Before you deploy#

• Your project is saved and valid. If it has custom code, the current revision needs a ready build (see Build).
• Your keys are stored. The banner reads API keys and secrets are ready, or says how many are still missing, with a shortcut to Open connections.
• You are an Organization admin or Operator. Developers can view deployments but cannot start them.

Deploy a service#

1. Where should it run?Choose a target. Local targets are good for testing; AWS targets create real cloud resources.
2. ConfigureEnter a Service name. Studio shows only the fields the target needs (see the table below).
3. Review the planSelect Preview plan. Studio lists every action it will take and any missing prerequisites. Select Refresh plan after you change something.
4. ConfirmTick I reviewed this plan, then select Deploy now.
5. Follow progressLog output appears as it runs. Select Cancel to stop it. When the service has an address, Open endpoint appears so you can reach it.

Run it yourself#

Prefer your own CI/CD? The box at the bottom of the Deploy tab downloads the project:

• Download bundle — for configuration-only projects: a portable bundle with the HTTP service, tests, environment template, container definitions, Compose, App Runner, and ECS templates.
• Download code — for projects with their own code: the code with its current settings in config/agent.yaml. Commit it to Git, import it into another Studio, or run it with algen-agent-runtime serve.

Members & roles#

Open Account menu›Organization settings›Members. Organization admins can invite people, change roles, resend or revoke pending invitations, and remove members.

Organization settings — Members (sample)

• 1Invite — email address plus the role they will join with.
• 2Members — change roles or remove people.
• 3Pending invitations — resend or revoke.

1. InviteEnter an email address, pick a role, and select Send invitation.
2. TrackPending invitations appear below active members, where you can resend or revoke them. Resending is rate-limited.
3. AdjustChange a member's role from the role selector, or remove them. An organization always keeps at least one admin.

Organization settings has five tabs: General (organization name and slug), Members, Integrations (GitHub and database hosts), Billing (subscriptions), and Security (sessions and linked identities). Select Back to Studio to return.

Roles#

Permission matrix#

Integrations#

Connect GitHub so your organization can import private repositories. Studio installs the Algen Studio GitHub App on your GitHub account or organization and reads only the repositories you grant. Each import gets a short-lived, read-only token for that one repository. Studio never stores a long-lived GitHub credential.

Organization settings — Integrations (sample)

• 1Connect GitHub — opens GitHub to install the app and choose repositories.
• 2Connected accounts — status, repository scope, a link to change repositories on GitHub, and Disconnect.

1. ConnectAn organization admin opens Settings › Integrations and selects Connect GitHub.
2. Grant accessOn GitHub, pick the account that owns the repositories and choose All repositories or only selected ones. You must be able to access that installation on GitHub.
3. ReturnGitHub sends you back to Studio, and the account appears as Active. Finish within 15 minutes of starting.
4. ImportAny member who can import projects can now import a granted repository by its URL.

To add or remove repositories later, select Manage repositories, which opens the installation on GitHub. You can connect more than one GitHub account. Disconnecting stops Studio from using that account for new imports; projects already imported are kept, and the app stays installed on GitHub until you uninstall it there.

A connection shows Suspended on GitHub or Uninstalled on GitHub when it was changed on GitHub. Reinstall or unsuspend it on GitHub, or disconnect it and connect again.

Database hosts#

The Integrations page also lists Database hosts. Agents and the Data tab reach only the databases declared here, read-only and through Studio's egress proxy. Use public DNS names; private addresses are never reachable.

An organization admin enters the Host, picks a Port, adds an optional Description, and selects Declare host. Other members see the list; if it is empty, ask an organization admin to add the host you need.

Credits & billing#

Hosted Studio meters the managed compute your runs, builds, and deployments use, and charges it against your organization's credit balance. Credits come from a recurring subscription, one-time credit packs, or both.

What is metered#

When a run starts, Studio reserves a conservative amount of credit. When it finishes, you are charged for actual usage — never more than the reservation — and the rest is released. If a run fails before using resources, the reservation is returned automatically.

AI credits for azAI#

In hosted Studio, azAI's own model calls use your organization's AI credits. Each answer uses credits for every step it takes. Before each call Studio reserves enough credit for it, then charges what it actually used and releases the rest. azAI's panel shows what a conversation has cost, and you can set a limit per conversation (see Conversations).

Manage billing#

Open Account menu›Billing and usage, or the Billing tab in Organization settings. Organization admins and billing admins can start a subscription:

1. Enter the plan codeUnder Start a subscription, enter the Plan code your AlgenAI account team gave you.
2. Check outSelect Continue to checkout. The secure hosted checkout opens in a new tab.
3. ConfirmBack in Studio, a card shows the subscription's status, plan, interval, and amount.

Organization settings — Billing (sample)

• 1Checkout — enter the plan code from your Algen account team.
• 2Subscription — status, plan, interval, and amount once the request is created.

When credits run out#

• New runs, builds, and deployments are not admitted and show an insufficient-credits message.
• Work already running finishes and is settled normally.
• Add credits or wait for your subscription renewal, then try again.

Your plan may also limit the number of projects, members, concurrent runs, queued runs, and requests per minute. These limits apply even when credits are available.

Security & sessions#

Open Organization settings›Security to review where you are signed in and which identities can sign in as you.

Organization settings — Security (sample)

• 1Sessions — revoke any browser you no longer use.
• 2Sign out everywhere — ends every session, including this one.
• 3Linked identities — unlink a provider while keeping at least one.

Active sessions#

Each session shows the browser and when it was last active. Your current session is marked Current. Revoke any browser or device you no longer use, or select Sign out everywhere to end every session, including this one.

Linked identities#

Your Google and GitHub sign-in identities are listed with their email and verification status. You can Unlink one as long as another remains. Unlinking signs you out of all sessions.

Keyboard shortcuts#

Press ? anywhere in Studio to see this list. On Windows and Linux, use Ctrl in place of ⌘.

Shortcuts dialog, opened with ? (sample)

• 1General — shortcuts that work anywhere in Studio.
• 2Go to — press G, then a letter, to jump to a section.

Single-letter shortcuts pause while you are typing in a field or a dialog is open.

In the Code tab#

Troubleshooting#

Glossary#

Admission

The check Studio performs before starting work: credits, concurrency, queue, and rate limits.

Agent

A model configured with instructions, tools, and limits to perform one job.

AI credits

The credits azAI's own model calls use in hosted Studio.

Approval

A point where a run pauses until a person approves, edits, or rejects what a step proposes. Compare clarification.

azAI

The assistant built into Studio. In Ask mode it explains; in Agent mode it makes changes after you allow them.

Build

A signed, scanned container image produced from a project revision.

Check

A run of the Code tab's stages — validate, lock packages, save version, build image, run tests, approval — that proves your code works.

Clarification

A point where a run pauses because it needs more information; you answer and it continues.

Credential

An encrypted secret value stored in Studio's vault and referenced by name.

Credits

Your organization's balance for metered managed compute.

Custom code project

A project that includes Python code and must be built before running in hosted Studio.

Custom logic

Python code in your project: custom workflow steps, hooks that prepare a request or check an answer, and code tools.

Data source

A database your agents may query read-only, set up on the Data tab.

Declarative project

A project defined entirely by configuration files.

Dependency lock

A file that pins every Python package to one exact version, so every build installs the same ones.

env:// reference

A pointer to a credential, such as env://OPENAI_API_KEY, used instead of a literal secret.

Evaluation suite

The set of test cases that checks an agent's behavior.

Hook

Custom code a workflow calls to prepare a request or check an answer. It must be approved before runs load it.

MCP

Model Context Protocol — a standard for connecting agents to external tool servers.

Organization

The workspace that owns projects, members, credits, and settings.

Revision

A saved version of a project. Each save creates a new revision.

SBOM

Software bill of materials — the list of components inside a build.

Template

A ready-made starting project, such as API checker or Human approval, offered in the New project dialog.

Test model

Studio's built-in model (provider mock, model deterministic). It is free and needs no key, but gives fixed answers.

Workflow

A pipeline of agents and steps, including optional human approvals.