Agents

Quick mental model: Chat answers. Agents act — in Gmail, Slack, your calendar, and the other apps you connect.

What Agent Mode Does

Agent mode lets the AI use connected apps as tools. Instead of only generating text, it can list messages, create calendar events, post to Slack, search GitHub issues, save files to Drive, and more — within the scope of your OAuth permissions.

Agent mode runs in Chat (switch the mode picker to Agent) or through Agent Recipes and Plan Mode steps that call tools.

Agent Mode vs Plan Mode vs Recipes

Plan Mode is the recommended successor to legacy Classic Pipelines for chained workflows.

Connect Apps

Before Agent mode can act, connect the apps you need:

1. Open Chat.
2. Click the Connected apps indicator (plug icon) near the composer.
3. Open Tools Marketplace — "Connect apps to supercharge your AI agent."
4. Pick an app category (Communication, Productivity, Development, CRM, etc.).
5. Complete OAuth in the popup.
6. Confirm status shows Connected.

The same connections power Live Data sources (Google Drive, OneDrive) on Contexts.

To disconnect, return to the marketplace and remove the connection.

Run an Agent in Chat

1. Connect the relevant apps.
2. Switch Chat mode to Agent.
3. Write a bounded instruction:

List my 5 most recent unread Gmail messages from today.
For each, write a one-line summary and a suggested reply tone (formal/casual).
Do not send anything — show drafts only.

1. Send and watch tool activity in the message (e.g. "Gmail · list_messages").
2. If OAuth expired, follow the re-authenticate prompt in the thread.

Writing good agent prompts

• Say what to read, what to create, and what to avoid.
• Ask for review before send on customer-facing actions.
• Name specific apps, folders, channels, or date ranges.
• Prefer small scoped tasks over open-ended "handle my email."

Agent Recipes

Create a Recipe with mode: Agent and select required toolkits. When someone runs it:

1. Springbase checks required apps are connected.
2. Variables are collected as usual.
3. Agent loop runs with your prompt template.

Example Agent Recipe prompt:

Using {{meeting_summary}}, post a concise recap to Slack channel {{channel_name}}.
Include action items with owners. Ask me to confirm before posting.

Plan Mode With Agent Steps

On Pro/Max, Plan Mode can include steps that invoke connected apps mid-plan — for example research in Chat, then create calendar holds, then draft email. Edit the plan before execution to remove steps you do not want.

Credits and Access

• Agent mode requires a paid plan (or account flag) and sufficient credits.
• Usage includes base model tokens plus per-tool charges where applicable.
• Running out of credits returns a clear error — top up in Settings → Billing.

Enterprise tenants may have:

• Agent feature policy disabled for members
• SSO enforcement — non-SSO sessions blocked on agent routes
• Rate limits per tenant on hot API paths

Popular Use Cases

Only prompt actions for apps you have connected.

Best Practices

1. Connect before you prompt — verify Connected apps shows green.
2. Confirm destructive actions — send email, post publicly, delete items.
3. Start read-only — list and summarize before write actions.
4. Save wins as Agent Recipes — same task next week in one click.
5. Use Contexts for knowledge — agent tools act; Contexts ground facts.

Troubleshooting

Related Guides

• Integrations — connection lifecycle and categories
• Recipes — Agent Recipes and variables
• Chat — mode picker and Context attachment
• Troubleshooting — broader fixes