Everything you need to scale autonomous agents
The full set of features behind building, running, and governing AI agents, broken down by category.
Building agents
- Drag & drop no-code builder: for agents, tools, workflows, evals and workforces
- Invent, our AI copilot: describe an agent and it wires the tools, knowledge, and evals
- MCP for coding agents: Claude Code, Codex and friends can build and manage your agents
- Sandboxed code execution (Python and Javascript)
- Version history with draft, active, and pinned states and one-click rollback
- Test, preview, and dry-runs with mocked outputs
Models
- All major LLMs: available out of the box
- Use via Relevance, or add your API keys
- BYO models
- Fallback models on failure: protect against vendor outages
- Switch model after N tokens (for cost optimization)
- Eval-driven model selection: use the cheapest model that does the job
Integrations & credentials
- 1,000+ pre-built app integrations
- Managed OAuth and API key flows
- Automatic OAuth token refresh and expiry handling
- Secret storage and credential vaulting per org and project
- Auth brokering: the model never sees keys or URLs
- Per-tool connected-account selection
- Custom MCP server registration: with auth-type config and custom headers
Knowledge & memory
- Full RAG ingestion pipeline: upload, parse, chunk, embed, index
- Hybrid retrieval (vector, keyword/BM25, and ensemble)
- Configurable chunking strategies (character, semantic)
- Managed vector store with per-project isolation
- Scheduled knowledge re-sync (Google Drive, Notion, Confluence, SharePoint)
- Long-term agent memory for preferences, facts, and instructions
- Observational memory and context compaction for long runs
- Tables for structured data agents read and write
- Custom metadata filled by agents
Orchestration
- Visual graph of your agent hierachy
- Forced handover and escalation edges mid-task
- "Agent-decide" dynamic routing
- Parallel agent execution with result merging
- Sub-agents exposed as callable tools with structured I/O
Triggers & channels
- Hundreds of pre-built triggers from events in your apps
- Trigger by API or webhook
- Scheduled and cron triggers (including agent self-scheduling)
- Build custom data signals as a trigger
- Inbound voice and phone-call agents
- Data-sync triggers (Gmail, Salesforce, HubSpot, Airtable, Jira, Notion, Drive)
- Slack and Teams channels with threaded replies
- Bulk trigger: run many agents over a dataset
- Debouncing, queueing and concurrency control
- Hosted chat app and mobile app
- Embeddable chat widget
Human oversight & approvals
- Per-action approval: required, not needed or up to agent
- Escalate-to-human tool with reason and context
- Bulk approve, reject, and respond
- Pause and resume of the queue
- Autonomy step limits: limit how deep an agent can go without oversight
- Cost-based pauses
- Terminate-on-limit
Evals
- Set up Checks and Test Suites to define performance expectations
- Pre-publish checks with thresholds
- Production sampling to catch drift and regressions
- Alerts if issues detected
- Eval-driven cost and performance optimization
Reliability & scale
- Durable execution and checkpointing
- Error-categorization framework (rate-limit, upstream, timeout, validation, auth, billing)
- Retry policy with exponential backoff
- Idempotency-aware retries
- Dead-letter queue
- Autoscaling workers
- Work-hours pacing
- Per-category error metrics and dashboards by integration
- Actionable error CTAs (reconnect account, top up credits)
Pre-built tool capabilities
- 1000s of pre-built integrations
- Image generation across 24+ models
- Browser automation
- Web scraping (Firecrawl, Apify, Browserless)
- Web search with many vendors and fallbacks
- File parsing and export to S3 (PDF, Word, Excel)
- Database query steps (Postgres, Snowflake, BigQuery, MongoDB, Supabase)
- Communication steps (email, Slack, Twilio SMS/voice, WhatsApp, Teams)
Enterprise security & governance
- SOC 2 Type II
- GDPR
- SSO and SAML
- Fine-grained RBAC across org, project, agent, tool, knowledge, and conversation
- SCIM directory sync
- Multi-region deployment and data residency, including a dedicated EU environment
- PII masking before data reaches the model
- Audit logging (user, IP, device, timestamp)
- Org, project, and data isolation
- Conversation-level sharing controls
Billing & multi-tenancy
- Usage-based billing and credits metering for your end-users
- Plan tiers and entitlements (seats, projects, knowledge limits)
- Per-org and per-project concurrency limits and queueing
- Overage and top-up flows
Observability & analytics
- Per-agent analytics (credits used, tasks processed, events)
- Conversation-level tracking (status, cost, tool stats, tags)
- Distributed tracing wired into logs and metrics
- Telemetry export via OpenTelemetry (OTEL)
- Delta Sharing for data export
- Concurrency and usage analytics per org and project
Sharing
- Marketplace to browse, clone, and submit agents, tools, and workforces
- Asset cloning with lineage tracking
- Public and embeddable agent publishing controls
