Base44 AI Agent
Base44 AI Agent
A privileged, skill-driven automation operator with cross-app read access and single-app write control. Build powerful workflows with a dumb trigger and a smart handler.
Core Mental Model
Dumb Trigger, Smart Handler
Simple events wake the agent. Once awake, it can do full logic, branching, chaining, API calls, emails, and write-backs.
Read Scope
BroadThe agent can read entity data from any app you own, using the app ID. Up to ~100 records per call with manual pagination.
Write Scope
LocalAll write actions are scoped to the attached app only. Schema changes, entity CRUD, and automations stay contained.
Capabilities
What the Agent Can Do
App Data
- Read entity data from attached app
- Read data from other owned apps
- Paginate through large datasets
- ~100 records per call
Schema & Entities
- Create & update entity schemas
- Add or remove fields
- Full entity CRUD in attached app
- Modify entity structure on the fly
Skills
- Write in JavaScript, Python, or Bash
- Run skills in sandbox environment
- Chain multiple skills together
- Persist skills in .agents/skills/
Automations
- Create automations in attached app
- Schedule-based triggers
- Entity event triggers
- Trigger skills and workflows
Files
- Read & write files in sandbox
- Parse uploaded or generated files
- Generate reports and exports
- Upload as public or private
External Services
- Browse & scrape the web
- Call public and authenticated APIs
- Use connected OAuth tokens
- LinkedIn, Google, Slack, Reddit…
Communication
- Recipient must be a registered app user
- Cannot email external addresses
- Generate images
- Log results to app entities
Memory
- Persist memory across sessions
- Save facts to memory files
- Reuse saved memory later
- Explicit save is most reliable
Boundaries
What the Agent Cannot Do
Cross-App Control
- Cannot modify another app's agent
- Cannot create schemas in another app
- Cannot create automations in another app
- Cannot write records into another app
Platform Access
- Cannot access Base44 billing
- Cannot access workspace settings
- Cannot access account settings
- Cannot see private dashboard internals
Security Admin
- Cannot modify RLS settings
- Cannot modify user role systems
- Cannot change access permissions
- Cannot harden your permission model
Runtime Boundaries
- Cannot create native backend functions
- Cannot email non-app users
- Cannot post as you without credentials
- Memory needs explicit save to persist
Important Warning — The agent operates with service-role style access inside its attached app. It may see all app data even if normal users cannot. Treat it like a privileged operator, not a normal user.
Use Cases
What to Build
Anomaly Flagging
Detect unusual activity, summarize trends on schedule, flag records for review.
Show examplesEvent-Driven Pipelines
Validate on create or update, enrich with computed fields, route to correct status.
Show examplesScoring & Classification
Score quality, urgency, or risk. Add reasons for scores. Standardize incoming data.
Show examplesSupport Triage
Categorize by issue type, detect urgency or frustration, suggest replies.
Show examplesContent Generation
Generate drafts from data, rewrite for different channels, create summaries.
Show examplesResearch & Enrichment
Browse public sources, pull facts from APIs, add missing context.
Show examplesScheduled Reports
Daily, weekly, monthly summaries. Highlight what changed. Management-friendly outputs.
Show examplesFile Processing
Extract text and data, parse structured files, generate exports.
Show examplesSmart Alerting
Add context to alerts, filter duplicate noise, route to right destination.
Show examplesExternal Integration
Authenticated API calls, sync with outside services, push results externally.
Show examplesData Auditing
Find missing or malformed data, detect duplicates, check workflow behavior.
Show examplesHuman-in-the-Loop
Highlight issues needing attention, prepare operator digests, reduce manual review work.
Show examplesStrong Patterns
Five Repeatable Patterns
Trigger → Analyze → Write Back
- Classification
- Validation
- Routing
- Scoring
Trigger → Analyze → Notify
- Support
- Monitoring
- Admin alerts
- Failure notifications
Schedule → Scan → Summarize
- Daily reports
- Weekly reviews
- Health checks
- Trend summaries
Input → Enrich → Store
- Leads
- Profiles
- Content metadata
- Research records
Upload → Parse → Structure
- CSV imports
- Reports
- Documents
- Logs
- Daily summary reports
- Ticket categorization
- Lead scoring
- File summarization
- Missing data audits
- Alerting on failed jobs
- Automatic schema changes
- Field removals
- Record deletions
- Bulk record modifications
- External posting without approval
Security Model
Trust & Access
- All entity data in attached app
- Secrets in .agents/.env
- Connected OAuth tokens
- Sandbox filesystem
- OAuth connectors
- Environment secrets
- Other app access via app ID
- Email sending to app users
- Other users' private apps
- Other workspaces' data
- Base44 billing or account settings
- Platform internals
Important Warning — The agent operates with service-role style access inside its attached app. It may see all app data even if normal users cannot. Treat it like a privileged operator, not a normal user.
Execution Limits
Know the Runtime Boundaries
Skill Timeout
Skills that run longer risk being interrupted. Break large jobs into smaller steps.
Records Per Call
Larger datasets require manual pagination — read, process, advance cursor, repeat.
Long-Running Jobs
Use an entity to track status, last cursor, error count, and final output.
Operating Rules
10 Rules for Developers
Treat the agent as a privileged system operator
Never allow schema edits without a preview
Add logging to every important skill
Add error handling to every important skill
Assume external APIs can fail at any time
Use entities to store job status, progress, errors, and output
Save important instructions explicitly to memory
Organize reusable skills cleanly in .agents/skills/
Review secrets and OAuth usage carefully
Test small before testing big