Agent Artifacts

Agent Artifacts enable agents to store large outputs (reports, analyses, datasets) for reuse, reducing redundant work and improving efficiency across your organization.

Overview

Without artifacts, agents face several challenges:

• Context bloat - Passing full research results through notifications bloats context
• Redundant work - The same research gets repeated for similar requests
• No discovery - Previous agent work is lost after execution completes
• Cost waste - Expensive LLM calls for identical queries

Artifacts solve these problems by providing a caching and storage layer for agent-generated content.

How Artifacts Work

Storage Flow

When an agent completes research or generates a report:

1. Content Creation - Agent produces valuable output (report, analysis, etc.)
2. Store Artifact - Agent calls store_artifact tool with content
3. Cache Key Generation - System generates SHA-256 hash for duplicate detection
4. Persistence - Artifact stored with metadata, tags, and validity period
5. Future Access - Other agents can search for and retrieve the artifact

Retrieval Flow

When an agent needs information that might exist:

1. Search Artifacts - Agent calls search_artifacts with relevant criteria
2. Results Returned - Matching artifacts with summaries are shown
3. Get Artifact - Agent retrieves full content using get_artifact
4. Usage Tracking - Access count and last accessed time updated

Deduplication

Artifacts use content hashing to prevent duplicates:

• SHA-256 hash generated from agent, type, title, and content preview
• If cache key matches an existing artifact, agents can reuse it
• Prevents storing the same research multiple times

Artifact Structure

Each artifact contains:

Field	Description	Example
Artifact ID	Unique identifier (UUID)	a1b2c3d4-...
Artifact Type	Category of content	report
Title	Short descriptive name	"Q4 Competitor Analysis"
Summary	Brief description of contents	"Analysis of 5 key competitors..."
Content	The full artifact content	Full report text
Content Format	Format of content	markdown
Tags	Comma-separated labels	"competitors, q4, market"
Category	Domain category	"marketing"
Is Reusable	Can other agents use it	true
Valid Until	Expiration date	2024-03-15
Usage Count	Times accessed	12
Created By Agent	Agent that created it	"Research Agent"

Artifact Types

Type	Description	Default TTL
report	Formal reports, summaries	30 days
analysis	Data analysis, insights	14 days
dataset	Structured data collections	7 days
document	General documents	90 days
data	Raw data, quick lookups	1 day

Choose the appropriate type based on content freshness requirements.

Content Formats

Format	Description	Best For
text	Plain text	Simple outputs
markdown	Formatted markdown	Reports, documentation
json	Structured JSON	Data, configurations
html	HTML content	Formatted content

Managing Artifacts

Viewing Artifacts

Navigate to Admin > Data & Logs > Artifacts.

The list shows:

• Title and type
• Creating agent
• Size and format
• Creation date and validity
• Usage count

Filtering Options

Filter	Description
Artifact Type	Filter by type (report, analysis, etc.)
Category	Filter by domain category
Status	Active, expired, superseded
Search	Search in title, summary, tags

Viewing Artifact Details

1. Click on an artifact in the list
2. View full content and metadata
3. See usage statistics and references

Artifact Statuses

Status	Description
active	Available for agent use
expired	Past validity date, read-only
superseded	Replaced by a newer version
archived	Manually archived by admin

Agent Tools

store_artifact

Stores agent findings as a reusable artifact.

Parameters:

Parameter	Required	Description
artifact_type	Yes	Type: report, analysis, dataset, document, data
title	Yes	Short, descriptive title
content	Yes	Full artifact content
summary	No	Brief summary
content_format	No	Format: text, markdown, json, html
tags	No	Comma-separated tags
category	No	Domain category
is_reusable	No	Allow other agents to use (default: true)
valid_for_days	No	Custom validity period

Example call:

{
  "artifact_type": "report",
  "title": "Q4 2024 Competitor Analysis",
  "summary": "Comprehensive analysis of 5 major competitors",
  "content": "[Full report content...]",
  "content_format": "markdown",
  "tags": "competitors, market, q4, 2024",
  "category": "marketing",
  "valid_for_days": 30
}

search_artifacts

Search for existing artifacts.

Parameters:

Parameter	Required	Description
artifact_type	No	Filter by type
category	No	Filter by category
tags	No	Filter by tags (comma-separated)
search_text	No	Search in title and summary
limit	No	Max results (default: 20, max: 100)

Example call:

{
  "search_text": "competitor analysis",
  "category": "marketing",
  "limit": 10
}

get_artifact

Retrieve full content of a specific artifact.

Parameters:

Parameter	Required	Description
artifact_id	Yes	UUID of the artifact
include_content	No	Include full content (default: true)

Enabling Artifact Tools

To allow agents to use artifacts:

1. Edit the agent in Configuration > AI Agents > Agents
2. In the Tools section, add:
   • store_artifact - Create new artifacts
   • search_artifacts - Find existing artifacts
   • get_artifact - Retrieve artifact content
3. Add instructions guiding artifact usage

Example Agent Instructions

## Artifact Usage Guidelines

Before performing extensive research:
1. Search for existing artifacts that might answer the question
2. If a relevant artifact exists and is still valid, use it instead of repeating work
3. If no artifact exists, perform the research and store valuable findings

When storing artifacts:
- Use descriptive titles that others can search for
- Include a summary explaining what's in the artifact
- Tag with relevant keywords for discoverability
- Choose the appropriate artifact type
- Consider if the content will remain relevant (set validity accordingly)

Artifact types:
- report: Formal summaries, research reports
- analysis: Data analysis, insights, comparisons
- dataset: Structured data, lists, tables
- document: General documents, templates
- data: Quick lookups, short-lived data

Use Cases

Research Caching

Scenario: Multiple customers ask about the same topic.

Without artifacts:

• Agent researches the topic each time
• Repeated API calls and LLM tokens
• Inconsistent answers

With artifacts:

• First request: Agent researches and stores artifact
• Subsequent requests: Agent finds and reuses artifact
• Consistent, cost-effective responses

Report Generation

Scenario: Weekly sales report requested by multiple stakeholders.

Title: Weekly Sales Report - Week 48 2024
Type: report
Summary: Sales performance metrics for Nov 25 - Dec 1, 2024
Tags: sales, weekly, november, 2024
Valid for: 7 days

Data Analysis

Scenario: Comparative analysis of product options.

Title: Enterprise Software Comparison - CRM Solutions
Type: analysis
Summary: Feature comparison of Salesforce, HubSpot, and Zoho CRM
Tags: crm, software, comparison, enterprise
Category: technology
Valid for: 30 days

Knowledge Extraction

Scenario: Agent processes a long document and extracts key points.

Title: Key Takeaways - Industry Report Q4 2024
Type: document
Summary: Summary of main findings from [Industry] annual report
Tags: industry, report, summary, 2024
Valid for: 90 days

Usage Tracking

Artifacts track access patterns:

Metric	Description
Usage Count	Total times the artifact was retrieved
Last Accessed	Most recent access timestamp
Created At	Original creation timestamp
Created By	Agent that created the artifact

Artifact References

The system also tracks which agents reference which artifacts:

• Reference Type: search_result, explicit_fetch
• Referencing Agent: Which agent accessed it
• Referencing Execution: Which execution accessed it

Use this data to:

• Identify high-value artifacts
• Understand cross-agent collaboration
• Audit information access

Best Practices

For Agents

1. Search before creating - Always check for existing artifacts first
2. Use descriptive titles - Make artifacts discoverable
3. Include summaries - Help other agents determine relevance
4. Tag consistently - Use established tag vocabulary
5. Set appropriate TTLs - Match validity to content freshness needs

For Administrators

1. Monitor usage - Review which artifacts are accessed most
2. Clean up expired - Remove or archive old artifacts
3. Establish conventions - Define standard categories and tags
4. Review periodically - Audit artifact quality and accuracy
5. Train agents well - Provide clear instructions on artifact use

TTL Guidelines

Content Type	Recommended TTL
Real-time data	1 day or less
Market analysis	7-14 days
Research reports	30 days
Reference documents	90 days
Timeless content	365 days or no expiration

Troubleshooting

Artifacts Not Being Reused

1. Check searchability - Are titles and tags descriptive?
2. Review agent instructions - Is the agent told to search first?
3. Verify validity - Has the artifact expired?
4. Check category/tags - Do search criteria match?

Duplicate Artifacts

If multiple similar artifacts exist:

1. Review agent instructions for clearer guidance
2. Consider consolidating into a single authoritative artifact
3. Archive or supersede older versions
4. Improve search criteria in agent prompts

Large Artifact Impact

For very large artifacts:

• Consider splitting into smaller, focused artifacts
• Use summaries to reduce retrieval overhead
• Monitor execution times when artifacts are loaded

Expired Artifacts

Expired artifacts are read-only but not deleted. To handle:

1. If still relevant, create a new artifact with updated content
2. If obsolete, leave expired (will be cleaned up per retention policy)
3. Update agent instructions if artifacts expire faster than expected

Security Considerations

Tenant Isolation

• Artifacts are scoped to the tenant
• Agents can only access their tenant's artifacts
• No cross-tenant artifact sharing

Content Sensitivity

• Artifacts may contain business-sensitive information
• Review stored content periodically
• Consider sensitivity when setting long TTLs
• Avoid storing credentials or PII in artifacts

Access Control

• Viewing artifacts requires settings:read permission
• Only agents with artifact tools can create/access artifacts
• All access is logged for audit purposes

Related Topics

• Personal Memories - Contact-specific information
• Global Memories - Organization-wide policies
• Tools - Artifact tool descriptions
• Agents - Agent configuration
• Agent Executions - View artifact usage in executions