Since the last update, we shipped 11 releases and 45 commits across the SDK. The headline: Agentuity Coder is here, bringing collaborative coding sessions to the platform. Let's get into it.

Overview

Agentuity Coder with a standalone SDK package, CLI command group, and interactive TUI for managing coding sessions
Claude plugin v2 with a streamlined skill-based architecture and a new deployment-focused project skill
New Explorer demos for databases, email, and queues: try Agentuity services hands-on without setting up a project
v2 stabilization with 10 patch releases hardening the new architecture
And more!

Agentuity Coder

Coder is now available as a standalone SDK package and CLI command group. It gives you a full lifecycle for collaborative coding sessions: create them, manage workspaces and skills, connect via the terminal, and run autonomous loops.

SDK Package

Install @agentuity/coder and start managing sessions programmatically:

import { CoderClient } from '@agentuity/coder';

const client = new CoderClient();

// Create a session with a task
const session = await client.createSession({
  task: 'Implement user authentication',
  workflowMode: 'standard',
});
console.log(`Session: ${session.sessionId}`);

// List active sessions
const { sessions } = await client.listSessions({ limit: 10 });
for (const s of sessions) {
  console.log(`${s.sessionId}: ${s.label} (${s.status})`);
}

The client resolves the Coder API URL automatically and authenticates with your AGENTUITY_SDK_KEY. You can also configure it explicitly:

const client = new CoderClient({
  apiKey: 'your-key',
  url: 'https://your-coder-url.example.com',
  orgId: 'your-org-id',
});

The full API surface covers sessions (create, get, update, list, delete, archive), workspaces, saved skills, skill buckets, GitHub account and repo integration, participant management, event history, and loop state.

CLI Commands

The agentuity coder command group covers the full session lifecycle:

# Create a session and connect the TUI
agentuity coder create "Build a REST API" --connect

# List sessions
agentuity coder list

# Launch the TUI for an existing session
agentuity coder tui --remote <session-id>

# Create a loop session for autonomous execution
agentuity coder create "Refactor auth module" \
  --workflow-mode loop \
  --loop-goal "Complete refactoring with tests" \
  --loop-max-iterations 20

Sessions support three visibility levels: private, organization, and collaborate. Loop mode lets sessions execute autonomously with configurable iteration limits and auto-continue behavior.

Claude Code Plugin v2

Our Claude plugin received a major overhaul. The previous multi-agent architecture has been replaced with a streamlined skill-based system that auto-activates based on what you're working on.

Five skills cover the full Agentuity workflow:

Skill	Activates When
`agentuity-project`	You want to deploy, host, or ship code
`agentuity-cloud`	You need databases, storage, queues, or other infrastructure
`agentuity-backend`	You're building with `@agentuity/runtime`, `@agentuity/schema`, etc.
`agentuity-frontend`	You're building with `@agentuity/react` or `@agentuity/auth`
`agentuity-ops`	You're running CLI commands or managing cloud resources

The new agentuity-project skill handles end-to-end project setup: creating Agentuity projects, restructuring existing apps for deployment, and managing infrastructure. The plugin also auto-updates your project's AGENTS.md when you use Agentuity services, so future sessions pick up where you left off.

New Explorer Demos

The SDK Explorer now includes interactive demos for three more services:

Database — Seed a products table and run type-safe queries with @agentuity/drizzle: filter by price, rating, or keyword, and compute aggregates against a live Postgres instance.

Email — Send a real email with ctx.email.send(), then poll for delivery status. You see the full send-to-delivery lifecycle in one run.

Queues — Create a queue, publish messages, and clean up. Shows ctx.queue.createQueue(), publish(), and deleteQueue() with retry and visibility timeout configuration.

Each demo runs in a cloud sandbox with real service access, so you're executing live code against live infrastructure, not reading static snippets. These join the existing set of demos covering KV, vector search, object storage, AI Gateway, streaming, and more. If you haven't tried the Explorer yet, we wrote about the thinking behind it in this walkthrough.

Also Shipping

Workflow Management

New agentuity cloud workflow list and agentuity cloud workflow get <id> commands for inspecting workflow state

Task Operations

agentuity cloud task close for batch-closing tasks with filters: status, type, priority, age, assignee, and more
--dry-run flag for previewing batch operations before executing
Enhanced agentuity cloud task list with --sort, --order, and filter flags

v2 Stabilization

Ten patch releases (v2.0.1 through v2.0.10) hardened the v2 architecture:

Fixed a Bun crash caused by drizzle-orm dependency chains
Gateway environment patching now runs before agent discovery in dev mode
Each agent gets its own isolated ctx.config from setup()
CDN asset URLs, analytics beacon, and Vite live tunnel hosts restored
Sandbox rm/rmdir now reports whether the target was found or already removed
start command renamed to tui (start and run remain as aliases)

v1 Maintenance

v1.0.64 shipped with sandbox job log streaming backported to v1, plus scaffolding and dist-tag fixes in create-agentuity.

Documentation

Redesigned docs landing page so it's easier to find what you're looking for, whether you're getting started, exploring services, or looking up reference docs
Explorer sandboxes granted services:read and services:write scopes for full service access in demos
Fixed KV API examples in the default template's AGENTS.md to match the current namespaced API
eval/evals renamed to evaluations throughout to avoid false positives in package security scanners

Upgrade Now

The CLI will prompt you automatically when a new version is available, or you can upgrade manually:

agentuity upgrade

New to Agentuity? Get started in seconds:

curl -fsSL https://agentuity.sh | sh

Once you're upgraded:

Install the Coder SDK: npm install @agentuity/coder
Create a coding session: agentuity coder create "My task" --connect
Try the new Explorer demos at agentuity.dev

Resources

Questions or feedback? Drop by our Discord to chat. Happy building!

Since the last update, we shipped 11 releases and 45 commits across the SDK. The headline: Agentuity Coder is here, bringing collaborative coding sessions to the platform. Let's get into it.

Overview

Agentuity Coder with a standalone SDK package, CLI command group, and interactive TUI for managing coding sessions
Claude plugin v2 with a streamlined skill-based architecture and a new deployment-focused project skill
New Explorer demos for databases, email, and queues: try Agentuity services hands-on without setting up a project
v2 stabilization with 10 patch releases hardening the new architecture
And more!

Agentuity Coder

SDK Package

Install @agentuity/coder and start managing sessions programmatically:

import { CoderClient } from '@agentuity/coder';

const client = new CoderClient();

// Create a session with a task
const session = await client.createSession({
  task: 'Implement user authentication',
  workflowMode: 'standard',
});
console.log(`Session: ${session.sessionId}`);

// List active sessions
const { sessions } = await client.listSessions({ limit: 10 });
for (const s of sessions) {
  console.log(`${s.sessionId}: ${s.label} (${s.status})`);
}

The client resolves the Coder API URL automatically and authenticates with your AGENTUITY_SDK_KEY. You can also configure it explicitly:

const client = new CoderClient({
  apiKey: 'your-key',
  url: 'https://your-coder-url.example.com',
  orgId: 'your-org-id',
});

CLI Commands

The agentuity coder command group covers the full session lifecycle:

# Create a session and connect the TUI
agentuity coder create "Build a REST API" --connect

# List sessions
agentuity coder list

# Launch the TUI for an existing session
agentuity coder tui --remote <session-id>

# Create a loop session for autonomous execution
agentuity coder create "Refactor auth module" \
  --workflow-mode loop \
  --loop-goal "Complete refactoring with tests" \
  --loop-max-iterations 20

Sessions support three visibility levels: private, organization, and collaborate. Loop mode lets sessions execute autonomously with configurable iteration limits and auto-continue behavior.

Claude Code Plugin v2

Our Claude plugin received a major overhaul. The previous multi-agent architecture has been replaced with a streamlined skill-based system that auto-activates based on what you're working on.

Five skills cover the full Agentuity workflow:

Skill	Activates When
`agentuity-project`	You want to deploy, host, or ship code
`agentuity-cloud`	You need databases, storage, queues, or other infrastructure
`agentuity-backend`	You're building with `@agentuity/runtime`, `@agentuity/schema`, etc.
`agentuity-frontend`	You're building with `@agentuity/react` or `@agentuity/auth`
`agentuity-ops`	You're running CLI commands or managing cloud resources

New Explorer Demos

The SDK Explorer now includes interactive demos for three more services:

Database — Seed a products table and run type-safe queries with @agentuity/drizzle: filter by price, rating, or keyword, and compute aggregates against a live Postgres instance.

Email — Send a real email with ctx.email.send(), then poll for delivery status. You see the full send-to-delivery lifecycle in one run.

Queues — Create a queue, publish messages, and clean up. Shows ctx.queue.createQueue(), publish(), and deleteQueue() with retry and visibility timeout configuration.

Also Shipping

Workflow Management

New agentuity cloud workflow list and agentuity cloud workflow get <id> commands for inspecting workflow state

Task Operations

agentuity cloud task close for batch-closing tasks with filters: status, type, priority, age, assignee, and more
--dry-run flag for previewing batch operations before executing
Enhanced agentuity cloud task list with --sort, --order, and filter flags

v2 Stabilization

Ten patch releases (v2.0.1 through v2.0.10) hardened the v2 architecture:

Fixed a Bun crash caused by drizzle-orm dependency chains
Gateway environment patching now runs before agent discovery in dev mode
Each agent gets its own isolated ctx.config from setup()
CDN asset URLs, analytics beacon, and Vite live tunnel hosts restored
Sandbox rm/rmdir now reports whether the target was found or already removed
start command renamed to tui (start and run remain as aliases)

v1 Maintenance

v1.0.64 shipped with sandbox job log streaming backported to v1, plus scaffolding and dist-tag fixes in create-agentuity.

Documentation

Redesigned docs landing page so it's easier to find what you're looking for, whether you're getting started, exploring services, or looking up reference docs
Explorer sandboxes granted services:read and services:write scopes for full service access in demos
Fixed KV API examples in the default template's AGENTS.md to match the current namespaced API
eval/evals renamed to evaluations throughout to avoid false positives in package security scanners

Upgrade Now

The CLI will prompt you automatically when a new version is available, or you can upgrade manually:

agentuity upgrade

New to Agentuity? Get started in seconds:

curl -fsSL https://agentuity.sh | sh

Once you're upgraded:

Install the Coder SDK: npm install @agentuity/coder
Create a coding session: agentuity coder create "My task" --connect
Try the new Explorer demos at agentuity.dev

Resources

Questions or feedback? Drop by our Discord to chat. Happy building!

April 3, 2026 Update

Overview

Agentuity Coder

SDK Package

CLI Commands

Claude Code Plugin v2

New Explorer Demos

Also Shipping

Upgrade Now

Resources

April 3, 2026 Update

Overview

Agentuity Coder

SDK Package

CLI Commands

Claude Code Plugin v2

New Explorer Demos

Also Shipping

Upgrade Now

Resources