Features
Everything your second team can do.
Configure once. They handle the rest. From multi-agent orchestration to enterprise security — every capability your AI workforce needs.
Multi-Agent Orchestration
A team that self-organizes.
Master agent classifies intent with LLM-based routing and delegates to specialized sub-agents running in parallel. Configurable concurrency, timeouts, and deterministic binding fallback.
- LLM-based intent classification with confidence thresholds
- Parallel sub-agent execution with configurable concurrency
- Deterministic binding router as fallback (priority rules, keywords, channels, roles)
- Session continuity — resume conversations, transfer between agents
- Execution monitoring with heartbeats and zombie detection
Built-in tools
They use the tools you already use.
Search, files, Git, browser tests, documents, and the systems your team already runs. An agent only receives the tools you allow.
- Core: web search, web fetch, calculator, datetime, memory, and scheduled jobs
- Files: list, read, write, and edit inside a sandboxed workspace. Parse Excel, PDF, Word, and PowerPoint.
- Code: Git, code search, Bitbucket pull requests, Playwright tests, and live previews
- Work systems: Slack, Jira, Notion, Figma, Pipedrive, Intercom, Startdeliver, and email
- Browser automation for sites that have no API
- Specialist agents cannot spawn further agents. Only the lead agent can.
MCP Protocol Support
The tool ecosystem that grows with you.
A curated catalog of MCP servers, with OAuth where the provider supports it. Use it for GitHub, databases, Google, and anything without a built-in tool — or bring your own server.
- Catalog includes GitHub, PostgreSQL, MongoDB, Google Drive, Google Workspace, Gmail, Sentry, Twilio, SendGrid, Mailgun, Brave Search, Vanta, and Heartpace HRM
- OAuth for Figma, Atlassian (Jira and Confluence), Notion, Intercom, Gmail, and Google Workspace
- Bring your own MCP server over stdio, SSE, or HTTP
- Tools from a new server show up without a product change
- Slack, Jira, Notion, Figma, Bitbucket, Pipedrive, Intercom, and Startdeliver also have built-in tools, separate from MCP
Knowledge Base & RAG
They know everything Dave knows.
Upload documents, auto-chunk, embed with pgvector. Your agents answer with source citations, grounded in your company's actual knowledge.
- Document ingestion with intelligent chunking pipeline
- pgvector embeddings with cosine similarity search
- Source citations in every response — [Source: document title]
- Per-agent knowledge scoping — each agent knows what it needs
- Similarity threshold filtering for high-quality results
Visual Workflow Builder
Automation without engineering.
Thirteen node types and five ways to start a run, plus a template library. Build the flow on a canvas, and pause when a person has to decide.
- Nodes: trigger, agent, tool, condition, channel, transform, loop, human input, delay, switch, HTTP, sub-workflow, and error handler
- Triggers: manual, schedule, webhook, channel message, and event
- Templates for engineering, sales, support, HR, finance, operations, and management
- Human input nodes stop the run until someone approves or answers
- You can watch each step as it runs
5 Channels
They show up where your team works.
Slack, Microsoft Teams, Email, Webhooks, REST API. No new app to roll out. Agents join the channels your people already use.
- Slack: socket mode bot with full message routing
- Microsoft Teams: Bot Framework adapter with webhook ingress
- Email: SendGrid, Mailgun, or SMTP. Inbound mail can be read. A reply is not sent until a person approves it.
- Webhooks: generic inbound/outbound with signature verification and rate limiting
- REST API: full API with JWT and API key auth, SSE streaming
LLM Provider Flexibility
Your models, your rules.
Anthropic, OpenAI, or your own local vLLM. Automatic fallback chains across providers. Per-tenant gateway policies for cost, privacy, and data residency.
- Anthropic, OpenAI, local vLLM (including Apple Silicon via vllm-mlx)
- Model resolution strategies: manual, policy, hybrid
- Automatic fallback chain — if one provider fails, the next picks up
- Per-tenant gateway settings for cost and privacy control
- Self-host with your own models for full data sovereignty
Built-in skills
Pre-built playbooks for every department.
Project management, code review, marketing, sales, incident response, design-to-code. Attach skills as operational playbooks to any agent.
- Playbooks across engineering, sales, marketing, support, HR, and ops — code review, incident response, pipeline review, design-to-code, RFI answers, and more
- Custom skills with prompt injection for your specific workflows
- Agent templates for quick setup — deploy a new agent in minutes
- Skills are composable — mix and match for each agent's role
Real-time Streaming & Observability
See everything they do.
SSE streaming with granular status events. Watch routing decisions, tool calls, sub-agent progress, and token usage in real-time.
- Granular SSE events: routing, thinking, tool calls, sub-agent progress
- Three chat modes: execute, plan, consult
- Prompt debug mode for full transparency into system prompts
- Sub-agent visibility panels showing parallel work
- Token usage and model information on every response
Enterprise Security & Governance
Better compliance than your first team.
Multi-tenant isolation at every layer. RBAC, tool policies, output redaction, audit logs. Your security team will approve this one.
- Multi-tenant isolation: HTTP guard, service layer, deterministic session keys
- RBAC with admin/manager/member roles and granular permissions
- Tool policies: allow/deny lists per agent and per binding
- Output redaction pipeline for sensitive data
- Security audit dashboard with severity scoring and remediation actions
- Full audit logging for every mutation
- API key authentication with SHA256 hashing
- Webhook signature verification and rate limiting
See it in action.
Book a demo and we'll show you how Team Two works for your use case.