Agents

​

What Are Agents?

• Adaptive: Different approaches for different situations
• Tool-driven: Use integrations and workflows as needed
• Conversational: Interactive back-and-forth discussions
• Context-aware: Consider your instructions and conversation history

​

Creating Your First Agent

You're a support operations assistant helping evaluate discount eligibility.

When given a Zendesk ticket number:
1. Retrieve ticket details and display them
2. Ask for confirmation before proceeding
3. Run "Get User Profile" workflow
4. Read discount criteria from Notion page [URL]
5. Evaluate eligibility with clear reasoning

Always respond professionally. If uncertain, ask for clarification.

Limiting Tool Capabilities (Recommended)

For MCP integrations, you can restrict which specific tools the agent can access. This makes your agent more reliable and prevents unintended actions.How to limit tools:

1. Click on any MCP integration you’ve added (Gmail, Salesforce, Slack, etc.)
2. Toggle off specific tools you don’t want the agent to use

Example configurations:

• Gmail: ✅ Search emails, Read emails → ❌ Send email, Delete email
• Salesforce: ✅ Get account, Search records → ❌ Delete record, Update record
• Slack: ✅ Read messages, Search channels → ❌ Send message, Create channel

Why this matters:

• Prevents destructive actions (deleting, sending)
• Makes agent behavior more predictable
• Reduces risk of unintended operations
• Improves reliability by limiting decision space

Add this to your system prompt: “If a user asks you to perform an action that’s disabled (like sending an email), explicitly tell them that capability is not available for this agent.”

​

How Agents Work

​

Tools (What Agents Can Use)

MCP Integrations

Direct connections to external services:

• Gmail: Read, search, and send emails
• Salesforce: Query records and update data
• Notion: Search documentation and databases
• Zendesk: Retrieve and manage support tickets
• Google Calendar: Check availability and schedule meetings
• And many more

Agents use the personal default credentials of whoever is running them (unless team credentials are configured). You’ll be prompted to authenticate if needed.

Gumloop Workflows

Your workflows become powerful tools for agents:

• Agents see the workflow name and description
• Understand expected inputs and outputs
• Call workflows when appropriate
• Process results to inform next steps

Example: A workflow that queries BigQuery, pulls Salesforce data, and combines them becomes a single “Get User Profile” tool for your agent.

Custom MCP Servers

Connect your own MCP server for services not available natively:

• Internal company APIs
• Custom data sources
• Specialized tools

Learn more in the MCP documentation.

Code Sandbox (Built-in)

The Code Sandbox is natively enabled on all agents. Your agent can execute code in a secure, isolated environment:

• Run Python code for data analysis, visualizations, and computations
• Execute shell commands for file operations and package installation
• Read/write files in the sandbox filesystem
• Upload/download files between Gumloop storage and the sandbox

The sandbox comes with 80+ pre-installed Python packages including pandas, numpy, matplotlib, scikit-learn, and more. See the Code Sandbox section below for details.

Web Search & Web Fetch (Built-in)

Agents have built-in Web Search and Web Fetch tools that are enabled by default. These allow your agent to search the web and fetch content from URLs without needing to connect a separate integration.

• Web Search: Search the web for information relevant to the current task
• Web Fetch: Retrieve and read content from specific URLs

You can configure the Fetch Provider (Firecrawl, Parallel, Exa, or Gumloop) to control how web content is retrieved. Use the three-dot menu to enable, disable, or change provider settings.

These tools are available out of the box — no API keys or integration setup required.

Search Past Conversations

When enabled, your agent can search and retrieve previous conversations to find relevant context from past interactions. This is a powerful tool for building agents that improve over time.How it works:

• Search by keyword: The agent searches past conversations for messages matching your query, ranked by relevance
• Retrieve full conversations: Once a relevant conversation is found, the agent can load the full message history for detailed context
• Date filtering: Narrow searches to specific time ranges (e.g., conversations from the last week)
• Team-wide search: For project/team agents, search across all team members’ conversations — not just your own

Why this matters:Search Past Conversations makes it easy to continuously improve your agent over time. For example, you can tell your agent:

“Look at the past 10 conversations and find gaps in how you handled requests. Then update your skills or system prompt based on the feedback I provided in those interactions.”

Because your agent can reflect on its own history, it can identify patterns, spot recurring issues, and proactively refine its behavior. This creates a feedback loop where every conversation makes the agent smarter and more aligned with how you work.Use cases:

• Self-improvement: Ask your agent to review past conversations and update its skills or instructions based on what it learns
• Context recall: “What did we discuss about the Q3 marketing plan?” — the agent finds and summarizes relevant past conversations
• Pattern detection: Identify recurring questions or issues across conversations to improve workflows
• Onboarding: New team members can ask the agent to surface relevant past discussions on a topic

Toggle this on from the Tools section of your agent configuration. It is found alongside other built-in tools like Code Sandbox and Skill Editing. Use the three-dot menu to enable or disable it.

Combine Search Past Conversations with Self-Improve Instructions and Skills to create an agent that genuinely learns from experience. The agent can search its history, identify what went well or poorly, and update its own instructions and skills accordingly.

Skill Editing & Creation

This toggle controls whether your agent can create and edit skills on its own during conversations. It’s found in the Tools section of your agent configuration and is enabled by default.

Enabled (default, recommended): Your agent can create new skills, update existing ones, and fix its own mistakes. This is the preferred option because it allows the agent to be iterative and continuously improve its playbooks over time.Disabled: Your agent can still read and use the skills you’ve already attached, but it cannot create new skills or edit existing ones on the fly. Use this if you want tighter control over your skill library.

Keep this enabled to get the most value from skills. An agent that can refine its own playbooks gets better with every conversation.

Tool Discovery (Custom Agents)

When an agent has many tools connected, loading every tool schema into context at once can eat into the model’s token budget. Tool Discovery solves this by lazy-loading tool schemas: instead of injecting every schema upfront, the agent starts with a lightweight list of available servers and tools, then pulls in the full schema only when it decides a tool is relevant to the current task.Why this matters:

• Keeps the agent’s context lean so more tokens are available for reasoning and conversation
• Lets you connect a large number of MCP servers without bloating every request
• The agent still has access to all connected tools, it just loads their schemas on demand

How to enable it: Tool Discovery is a toggle in the Abilities section of your agent’s configuration. Once enabled, the agent will lazy-load tool schemas from all connected MCP servers (guMCP, Gumstack, and custom MCP) for the user who is chatting, filtered by any restrictions you’ve set.

The general agent already uses Tool Discovery by default. This setting is specifically for custom agents where you want to keep context lean while still having broad tool access.

​

Instructions (How Agents Behave)

When the user provides a ticket number:
1. Retrieve ticket details and ask for confirmation
2. Once confirmed, run the "Get User Profile" workflow
3. Read the discount policy from Notion
4. Assess eligibility and explain your reasoning

​

Self-Improve Instructions

​

Reasoning (How Agents Think)

1. Analyzes the request and available tools
2. Decides which tools to use and in what order
3. Executes tool calls and adapts based on results
4. Asks for confirmation when needed (based on instructions)
5. Explains its reasoning step-by-step

​

Triggers

​

Skills

• Encode multi-step processes the agent should follow every time (outreach sequences, triage checklists, reporting workflows)
• Store templates and domain knowledge too detailed for the system prompt
• Load only when relevant, saving tokens compared to stuffing everything into the system prompt
• Improve over time as the agent learns from your feedback and even creates new skills on its own

​

Embedding Agents in Workflows

​

Using Workflows as Agent Tools

Input Node: "company_name" (text)
Input Node: "date_range" (text)
[Your workflow logic here]
Output Node: "enriched_data" (object)
Output Node: "success" (boolean)

​

Code Sandbox

• Run Python code for data analysis, visualizations, and computations
• Execute shell commands for file operations and package installation
• Read and write files in the sandbox filesystem
• Upload and download files between Gumloop storage and the sandbox

​

Voice Input

​

How It Works

1. Record or upload an audio file in the chat input
2. Gumloop transcribes the audio using an AI transcription model
3. The transcribed text appears as your message in the conversation
4. The agent responds to the transcribed text like any other message

​

Supported Formats and Limits

​

Transcription Models

​

Subagents

​

How Subagents Work

​

Configuring Subagents

​

Behind the Scenes

1. Interaction created: Each subagent gets its own interaction record. The parent agent’s conversation references the child interactions, and all subagent conversations are visible in the chat history.
2. Parallel execution: Multiple subagents run concurrently. The number of simultaneous subagents is based on your subscription tier.
3. Progress tracking: For batch invocations (multiple subagents at once), a shared progress board tracks each child’s status. Sibling agents can see each other’s progress and communicate via broadcast notes.
4. Artifact transfer: The parent can transfer specific files from its sandbox to a subagent before execution. After the subagent finishes, its conversation transcript is saved and can be read by the parent.
5. Timeout handling: Subagents run as queued background tasks with their own execution budget (approximately half the parent agent’s time limit). If a subagent exceeds its time limit, it receives a graceful abort signal followed by a hard cancellation if it does not stop.
6. Depth limits: Self-cloning has a depth limit of 1, meaning a clone cannot clone itself again. However, cross-agent chains (Agent A invokes Agent B, which invokes Agent C) are unlimited.

​

Subagent FAQ

​

Message Queue and Steering

​

How It Works

1. While the agent shows “Responding…”, type and send additional messages as usual
2. Your messages appear in a queue above the input field, in the order you sent them
3. Between processing steps (for example, after the agent finishes a tool call), the agent automatically drains the next queued message and incorporates it into the conversation
4. The agent treats the queued message as if you had sent it at that moment, adjusting its response accordingly

​

What You Can Do With Queued Messages

​

Steering

• Redirect the agent: “Actually, focus on the Q3 numbers instead”
• Add missing context: “By the way, the deadline is Friday”
• Refine the request: “Make it more concise” or “Include a chart”
• Stack instructions: Queue up a sequence of tasks for the agent to work through

​

AI Advanced Settings

• Model Config
• Auto Summarization
• Fallback

Customize AI model behavior. Available parameters vary by provider (OpenAI, Anthropic, Google). Settings are stored per-provider, so switching models preserves your preferences for each.

OpenAI Parameters

Parameter	Range	Default	Description
Reasoning Effort	low, medium, high	medium	Controls computational effort before responding. Higher = more thorough reasoning but slower and more tokens. Only for o-series and GPT-5 models.
Temperature	0–2	1	Controls output randomness. 0 = deterministic, 2 = highly creative.
Max Output Tokens	1+	Auto	Upper bound for generated tokens (includes reasoning tokens). If unset, uses model default.
Max Tool Calls	1+	Auto	Limits total tool calls per response. Useful for cost control.
Top P	0–1	1	Nucleus sampling. Model considers only tokens in the top P probability mass. Only for GPT-4o, GPT-4.1, GPT-5.1, GPT-5.2, o3, o4.
Parallel Tool Calls	on/off	on	When on, model can execute multiple tools simultaneously. Disable if tools depend on each other’s results.

Adjust either Temperature or Top P, not both. Using both produces unpredictable results.

Anthropic (Claude) Parameters

Parameter	Range	Default	Description
Extended Thinking	enabled, disabled	enabled	Shows Claude’s reasoning process before the final answer. When enabled, temperature is forced to 1.0.
Budget Tokens	1,024+	10,000	Token budget for thinking. Higher = more thorough reasoning but slower. Only shown when thinking is enabled. Must be less than Max Tokens.
Temperature	0–1	1	Controls output randomness. Note: forced to 1.0 when Extended Thinking is enabled.
Max Tokens	1+	Auto	Maximum tokens to generate. Defaults: 64,000 (Claude 4.5), 8,192 (older models).
Top P	0–1	1	Nucleus sampling threshold.
Top K	1+	Auto	Limits sampling to top K most likely tokens. Lower values (10–40) produce more focused outputs by removing long-tail responses.
Disable Parallel Tool Use	on/off	off	When on, model outputs at most one tool call per response. Enable for strict sequential execution.

Google (Gemini) Parameters

Parameter	Range	Default	Description
Thinking Level	LOW, HIGH	HIGH	Controls depth of internal reasoning. LOW = faster responses, HIGH = deeper analysis. Only for Gemini 2.5 and 3 models.
Temperature	0–2	1	Controls output randomness.
Top P	0–1	0.95	Nucleus sampling. Note: Google’s default (0.95) is lower than other providers.
Top K	1+	Auto	Maximum tokens considered at each generation step.
Max Output Tokens	1+	Auto	Maximum tokens to generate.

Temperature vs Reasoning: Temperature controls randomness in word selection. Reasoning effort/thinking controls how much the model deliberates before responding. For complex analysis, increase reasoning, not temperature.

Manages long conversations by summarizing older messages when context limits are reached.

When a conversation approaches the model’s context limit, the system identifies the oldest messages, generates a structured summary (Goal, Actions Taken, Key Data, Status, Next Steps), and replaces them while keeping recent messages in full detail.

Compaction operates at the part level within messages, not just message level. This enables finer-grained compaction when a single message contains many large parts (e.g., 24 tool call results totaling 150k+ tokens).

Behavior	Auto Mode (Default)	Override Mode
When to summarize	System determines automatically	System determines automatically
Summary model	Auto-selected (fast, cost-effective)	You choose the model
Prune Protection	Optimized default (40,000 tokens)	Configurable
Summary Max Tokens	Optimized default (30,000 tokens)	Configurable

Keep defaults unless you have specific context requirements. Lower Prune Protect = more aggressive summarization. Higher Summary Max = summaries trigger earlier.

Ensures reliability by automatically switching to alternative AI models when your primary model is unavailable.

When errors occur, the system classifies the error, retries based on severity, then falls back to the next model in the chain. Models from the same provider are automatically excluded for true redundancy.

Error Type	Retries Before Fallback	Rationale
Rate Limit	2	Rate limits often clear quickly
Provider 5xx	1	Provider unhealthy, fallback faster
Network Error	0	Immediate fallback (no retries)
Timeout	1	Server overloaded, try once then fallback

Auto Mode (Default): Fallback models are selected based on your primary model’s characteristics:

Primary Model Type	Fallback Chain
Expert/Smartest (intelligence rating > 4)	Claude Opus 4.5 → Gemini 3 Pro → GPT-5.2
Fastest (speed rating > 4)	Gemini 3 Flash → Claude Haiku 4.5 → GPT-4.1
Recommended/Balanced	Claude Sonnet 4.5 → Gemini 3 Flash → GPT-5.2

Override Mode: Manually select up to 2 fallback models. Drag-and-drop to set priority order. Models from the same provider as your primary are automatically filtered out.

Disabling fallback means your agent will fail if the primary model is unavailable. Keep enabled unless you have specific compliance requirements.

​

Context Usage Meter

​

Understanding Credit Costs

​

Model Pricing Overview

​

Workflow and Integration Costs

• Free: Google Workspace, Slack, Airtable, Notion, Salesforce, HubSpot, 150+ more
• Paid: Data enrichment services (eg. Hunter.io ~10 credits, ZoomInfo ~60 credits)

​

Optimizing Credit Usage

​

Tracking Credit Usage

• Real-time Display: Credits shown next to each agent response as it streams
• Conversation History: View total credits per conversation thread
• History Page: See credit costs for agent chats and Agent node executions (when run via workflows)
• Account Dashboard: Track total credits used across all agents and timeframes

​

Credentials & Authentication

• Use Personal Default — Each person who runs the agent uses their own default account. This is the default.
• Use Team Default (team agents only) — Everyone on the team uses the same shared account.
• Use Specific Account — Pin a specific account for this agent. Useful when you have multiple accounts for the same service.

​

App Rules

​

Agent-Level Rules

​

Best Practices & Troubleshooting

Never:
- Delete customer data or records
- Send emails without explicit user approval
- Make purchases or financial commitments
- Override manual decisions by team members
- Modify production data without confirmation

​

Finding Agents

​

Managing Chats

​

General Preferences

​

Follow-Up Prompts

​

Incognito Mode

​

How to Use Incognito Mode

​

What Happens in Incognito Mode

​

Next Steps