Changelog

We're excited to announce two powerful new tools for working with PDFs: a dedicated PDF Search & Replace skill and a ready-to-use Brad - PDF anonymization agent template. Together, they make it easy to find and replace text while protecting sensitive data.

PDF Search & Replace skill

Find and replace text in PDF files while keeping your original formatting intact. Whether you're updating placeholders, correcting information, or preparing documents for distribution, this skill handles the work automatically.

Provide the PDF URL, specify the strings to search for and their replacements, and the processed PDF is uploaded to cloud storage with a download link returned instantly.

• Available on all plans

• No file size limits

• Processed PDFs uploaded to S3 with instant download links

• Preserves original PDF formatting during replacements

Where to find it: Skills → Search "pdf" → PDF Search & Replace

How to test it: Add the skill to any agent, then ask it to search and replace text in a sample PDF. For example: "Search for [PLACEHOLDER] in this PDF and replace it with [ACTUAL_VALUE], then give me the download link."

Brad - PDF anonymization agent template

Protect sensitive information automatically. Brad is a pre-configured agent template that detects and removes all personally identifiable information (PII) from PDF documents in seconds.

Upload a PDF containing names, contact details, financial information, or any other sensitive data. Brad analyzes the document, identifies all PII, removes it, and returns a clean, anonymized PDF ready to share or archive.

• Detects all types of PII (names, emails, phone numbers, addresses, financial data, and more)

• Available on all plans

• No file size limits

• Returns anonymized PDF as a downloadable file

• Ready-to-use template — no configuration needed

Where to find it: Agents → Create Agent → From Template → Brad - PDF Anonymization

How to test it:

1. Go to Agents and click "Create Agent"

2. Select "From Template"

3. Search for and select "Brad - PDF Anonymization"

4. Choose "Clone and test" to use it immediately

5. Upload a PDF with sensitive information

6. Brad will process it and return the anonymized version as a download

Why this matters

These tools solve a real problem: managing sensitive documents at scale. Whether you're preparing contracts for external review, sharing reports while protecting privacy, or automating document workflows, you now have enterprise-grade PDF handling built into your agent fleet.

Available now on all Swiftask plans. No additional setup or configuration required.

Ready to try it? Create a new agent from the Brad template or add the PDF Search & Replace skill to your existing agents. Both are available immediately in your workspace.

Looking for the perfect model for your AI agents? We've just added two new Claude models via AWS Bedrock: a champion of value for money, and a beast of raw power.

Claude Sonnet 4.6 — Advanced intelligence without compromising your costs

You get near-Opus power at the same price as Sonnet 4.5. Perfect for agents that demand serious reasoning without breaking the bank.

• Pricing: 3/3 /3/15 per million tokens (input / output)

• Capabilities: Extended thinking, web automation (computer use), image analysis, up to 1M token context

• Ideal for: Complex autonomous agents, full code generation, strategic analysis of large documents

• Key benefit: Cut your AI infrastructure costs by 40% compared to Opus, without sacrificing quality

Claude Opus 4.6 — When you need the best

Anthropic's most powerful model. Use it when reasoning quality is non-negotiable and you're tackling critical or highly complex tasks.

• Capabilities: Maximum reasoning, ultra-precise instruction following, handling complex edge cases

• Ideal for: Critical tasks, multi-step problems, analyses where precision makes the difference

• Key benefit: Zero compromise on quality — Anthropic's most capable model

How to use these models

In your agents:

1. Go to Agents → Select your agent → Edit

2. Open the Objectives & Instructions section (or Model selection)

3. Click the LLM Model dropdown

4. Search for and select Bedrock – Claude Sonnet 4.6 or Bedrock – Claude Opus 4.6

5. Save

In Chat:

1. Open Chat

2. Click the model selector (top left of the input field)

3. Search for the model by name

4. Start chatting

Availability

• ✅ Available on all Swiftask plans (Pro, Team Starter, Team Growth, Tailored)

• ✅ No upgrade required

• ✅ All your existing agents, skills, and automations work immediately

Our advice

Start with Sonnet 4.6 for your standard agent workflows — it's the best balance of power and cost. Reserve Opus 4.6 for your critical tasks where maximum reasoning precision makes the difference. This way, you optimize your budget while keeping the power where it matters most.

How to test

Claude Sonnet 4.6:

• Create an agent with a complex task: code analysis, strategic synthesis, or architecture generation

• Example: "Analyze this full financial report and extract the 10 most important insights"

• Test multimodal capabilities: Upload an image and request a detailed analysis

Claude Opus 4.6:

• Use it on a critical task where reasoning quality is essential

• Example: "Identify all edge cases and error scenarios in this process flow"

• Compare results with Sonnet 4.6 to see the difference in precision

Ready to deploy your agents with the best models? Open your agents now, select Claude Sonnet 4.6 or Opus 4.6, and test the power. 🎯

We've added a powerful new way to keep data private and accountability clear when sharing agents with your team. Meet the "Require user's own account" toggle—a simple but important security feature for external integrations.

Agents now respect user credentials for external services

When you share an agent that uses external integrations (Gmail, Google Drive, Slack, Outlook, Microsoft Teams, or any calendar service), you can now require users to authenticate with their own accounts instead of using yours.

Here's what changes:

• Each user's data stays on their account – When someone uses a Gmail skill, they use their Gmail account, not yours. Their emails, files, and actions are completely separate.

• Clear accountability – Every action is traceable to the person who performed it, not to the agent owner.

• Automatic permission respect – Users can only access what their own account permissions allow. No surprises, no over-sharing.

• One-time setup – Users configure their account once when they first access the agent, then everything works seamlessly.

How to set it up

1. Navigate to Agents → select your agent → click the edit icon (✏️)

2. Go to Skills in the left panel

3. Click Add a skill or edit an existing external skill (Gmail, Google Drive, Slack, Outlook, Calendar, etc.)

4. Fill in the skill configuration form

5. At the bottom of the form, toggle "Require user's own account" on

6. Save and share your agent normally

That's it. When recipients access the agent, they'll see a "Configuration required" banner and can authenticate with one click.

Available on: All Swiftask plans
Works with: Gmail, Google Drive, Slack, Outlook, Microsoft Teams, Google Calendar, Outlook Calendar, and all external service integrations
Does not apply to: Internal Swiftask skills (only external integrations)

What your users will see

1. They open a shared agent that has this toggle enabled

2. A blue "Configuration required" banner appears

3. They click "Configure" and authenticate with their own account

4. They're done—the agent now uses their credentials for all external actions

Why this matters

• Privacy first – Personal data (emails, files, calendar events) stays on the user's account

• Compliance ready – Perfect for regulated industries where audit trails and individual accountability are essential

• No credential sharing – You never have to share your Gmail, Slack, or calendar access with teammates

• Respects existing access – If someone doesn't have permission to a resource in their own account, they can't access it through the agent either

Test it now

1. Create or edit an agent with an external skill (try Gmail or Google Drive)

2. Enable the "Require user's own account" toggle

3. Share the agent with a test user (different account)

4. Log in as the test user and open the shared agent

5. Look for the "Configuration required" banner

6. Click "Configure" and authenticate

7. Send a test message that triggers the skill

8. Verify the action used the test user's credentials, not yours

Ready to share securely? Open any agent, add an external skill, enable the toggle, and share with confidence. Your team's data stays where it belongs—with them. 🔒

We've added agent locking so you can protect your production agents from accidental changes. Once locked, only owners and admins can modify agent settings—everyone else can still use the agent, but they'll see a clear message if they try to edit it.

Lock agent in production

You can now lock any agent you own or administer to prevent unauthorized changes. Perfect for production agents, shared templates, or any agent you want to keep stable. When locked, all configuration changes are blocked for regular users—they'll see a helpful error message asking them to contact the owner or admin.

• Turn the lock on or off with a single toggle

• Only owners and admins can lock or unlock agents

• Locked agents remain fully usable in chat for everyone with access

• All agent settings are protected: prompt, skills, data sources, sharing, duplication, and more

• No impact on usage, credits, or performance

Find it: Agent page → top section → "Lock agent" toggle (next to Share and "Use this agent" buttons)

Testing the feature:

1. Open any agent you own or administer

2. Click the "Lock agent" toggle at the top of the page (it will turn on and stay highlighted)

3. Try to edit the agent settings as a regular user—you'll see a message asking to contact the owner/admin

4. As owner/admin, click the toggle again to unlock and make changes

We've just made Swiftask smarter. OpenAI's GPT-5.3 is now available across all plans, bringing cutting-edge reasoning and logic capabilities to every team member. No upgrades. No extra costs. Just better thinking.

GPT-5.3 — Advanced reasoning for your toughest problems

GPT-5.3 is the most advanced model in the GPT-5 family. It delivers superior logic and coding capabilities, with deep reasoning and complex instruction following built in. Whether you're solving intricate problems, writing sophisticated code, or working through multi-step logic, GPT-5.3 gets it right.

Why it matters: This is the highest level of intelligence available right now. Use it when you need your AI to think harder, reason deeper, and handle complexity that other models struggle with.

• Advanced reasoning – Tackles multi-step problems with precision

• Superior logic & coding – Writes better code, catches edge cases, explains complex logic

• Deep reasoning & complex instruction following – Understands nuanced requirements and delivers accordingly

• Performance – Outperforms GPT-4o on MMLU and GSM8K benchmarks

• Available on all plans – Pro, Team Starter, Team Growth, and Custom at no extra cost

How to use GPT-5.3

In Chat:

1. Open Chat in the left sidebar

2. Click the AI model selector (top left, showing your current model)

3. Search for or select "OpenAI GPT-5.3" from the dropdown

4. Start your conversation — GPT-5.3 is ready to go

Set it as your default:

1. Go to Workspace administration → Settings → Advanced

2. Find Default Chat AI

3. Select "OpenAI GPT-5.3"

4. Click Update

All new conversations will now use GPT-5.3 by default.

How to test it

Try GPT-5.3 on tasks where reasoning matters most:

• Code review – Ask it to analyze complex code, identify bugs, suggest optimizations

• Research & analysis – Complex document analysis, data synthesis, strategic thinking

• Problem-solving – Multi-step logic, edge case handling, constraint satisfaction

• Writing & explanation – Nuanced writing, technical explanations, instruction generation

Start a conversation with a challenging prompt and see the difference advanced reasoning makes.

Google's latest Gemini models are now available across all Swiftask plans. Whether you need lightning-fast responses or deeper reasoning, you can now choose the right Gemini model for every task — no plan upgrades needed.

Gemini Pro 3.1 — Peak performance for complex thinking

You can now access Gemini Pro 3.1 for tasks that demand sophisticated reasoning and nuanced analysis. This model excels at breaking down complex problems, generating detailed strategies, and handling multi-step workflows with precision.

• Better reasoning capabilities for complex, multi-step tasks

• Faster inference than previous versions

• Available on all plans — Pro, Team Starter, Team Growth, and Custom

• Configure as default in Workspace administration → Settings → Advanced

How to test it:

1. Navigate to Chat

2. Click the AI model selector at the top

3. Search for "Gemini Pro 3.1" and select it

4. Try a complex reasoning task: "Break down a go-to-market strategy for a new SaaS product, including positioning, target segments, and 90-day milestones"

5. Compare response depth and clarity with previous Gemini versions

Gemini 3.1 Pro — Speed meets intelligence

Gemini 3.1 Pro brings you faster inference without sacrificing reasoning quality. Perfect for high-volume tasks, real-time interactions, and workflows where speed matters as much as accuracy.

• Faster inference for quicker responses

• Strong reasoning for quality outputs

• Available on all plans — Pro, Team Starter, Team Growth, and Custom

• Ideal for chat interactions, meeting transcription, and artifact generation under time constraints

How to test it:

1. Open Chat

2. Select the AI model dropdown

3. Choose "Gemini 3.1 Pro"

4. Try a time-sensitive task: "Generate a meeting summary with 5 action items from this transcript [paste content]"

5. Compare speed and output quality

How to switch between models

Both models are available everywhere you use AI in Swiftask:

• In Chat: Click the model selector at the top of the input field

• As workspace defaults: Workspace administration → Settings → Advanced → Default Chat AI (or Meeting AI, Transcription AI, Artifact AI)

• For agents: Agent configuration → Model selection

All models are included in your current plan. No upgrades needed.

Improvements

Real-time model recommendations now surface Gemini Pro 3.1 and Gemini 3.1 Pro as options based on your task description, helping you pick the right tool instantly without guessing.

Ready to experience faster, smarter AI? Open Chat now and switch to Gemini Pro 3.1 or Gemini 3.1 Pro to see the difference.

We've given sub agents their own dedicated space in your agent settings, making it easier to visualize and manage multi-agent workflows.

Sub agents: visualize your agent network

You can now see exactly how your agents are connected and what workflows they belong to, right from your agent configuration. The new sub agents panel shows a visual map of your agent architecture, helping you understand your multi-agent system at a glance. Attaching new sub agents is just one click away.

What you can do:

• View a visual diagram showing how agents are interconnected

• See the full workflow each agent belongs to

• Attach new sub agents directly from the panel using the "Attach agent" button

• Support nested sub agents (agents with sub agents that have their own sub agents)

• Manage sub agents from the Skills section (view or delete)

Where to find it:

Agents → Edit agent → Sub agents (in the left sidebar)

Available on: All Swiftask plans

How to test it:

1. Navigate to Agents and select an agent to edit

2. Look for Sub agents in the left sidebar

3. Click the "+ Attach agent" button

4. Select an agent to attach as a sub agent

5. Verify the visual diagram appears showing the connection between agents

6. Try attaching multiple sub agents to create a more complex workflow

7. Test nested sub agents by attaching an agent that itself has sub agents

8. Delete a sub agent from the Skills section to confirm management works smoothly

We know how crucial it is to keep your AI agents running smoothly without unexpected hiccups. This month, we added agent monitoring with credit alerts to help you stay in control of your AI usage across your public fleet.

Agent credit monitoring

You can now set up automatic alerts when your public agents near their credit limits, helping you prevent service disruptions and manage your AI spend more effectively.

• Credit threshold alerts: Get in-app notifications when an agent's credit usage hits your set limit

• Custom thresholds: Set any credit limit by simply entering the number in the dedicated field

• Individual monitoring: Track usage limits for each agent separately

• Proactive notifications: Alerts kick in as usage approaches your threshold, not after it's exceeded

• For public agents only: Available for widgets and agents as apps (agent as app)

• Available on all plans with no usage limits

How to test it: Head to a public agent (widget or agent as app) → More options → Agent monitoring (located between Changelog and Advanced settings). Toggle on "Credit threshold alerts" and set your desired credit threshold.

Find it in: Agent settings → More options → Agent monitoring