Deploy AI Workflows
In Your Cloud
Grove is the enterprise orchestration platform for production AI. Build multi-step workflows as DAGs, run autonomous agents in sandboxed workspaces, route across LLM providers — all in your own infrastructure, with first-class multi-tenancy and the durability and auditability your organization demands.
Built for Enterprise AI
Your Cloud, Your Control
Deploy in your own infrastructure. No data leaves your environment. Air-gap compatible for the most sensitive workloads.
DAG-Based Orchestration
Define workflows as directed acyclic graphs. Automatic parallel execution, fan-in/fan-out, conditional routing, and bounded refinement loops for generator-critic patterns.
Autonomous Agents
Goal-driven loops with allowlisted tools, per-run sandboxed workspaces, durable turn history, and budget caps. Sub-agent delegation with depth and fan-out limits. Resume from any crash.
Multi-Provider LLM Routing
Anthropic Claude, OpenAI, Google Gemini, Azure OpenAI, and Vertex AI in a single workflow. Named model groups with automatic failover. Route per node. Prompt caching and per-token cost accounting keep spend visible and low. No vendor lock-in.
Multi-Tenant by Design
First-class tenant entity with Postgres row-level isolation. Per-tenant secrets, LLM credentials, quotas, rate limits, and budget caps. Tenant lifecycle API with provision-and-key in one call.
Production Durability
PostgreSQL-backed crash recovery. Resume from the last checkpoint. Every execution persisted and auditable. Per-tool-call idempotency keeps side-effecting work at-most-once across crashes.
How Grove Works
Define workflows as DAGs or autonomous agents. Grove handles execution, parallelism, streaming, durability, and tenant isolation — in your infrastructure.
Everything You Need for Production AI
External Tool Execution
LLM workflows pause while your application executes tools locally. Credentials never leave your environment. Persisted pending calls survive crashes.
Crash Recovery
PostgreSQL-backed checkpointing. Workflow runs resume from the last completed node; agent runs resume mid-turn with per-tool-call idempotency markers that keep side-effecting work at-most-once.
Real-Time Streaming
SSE event streams deliver node-by-node progress to your UI. Live agent-run replay across replicas. No polling required.
Encrypted Secrets
AES-256-GCM encryption at rest, plus a pluggable backend that can delegate to a HashiCorp Vault KV v2 mount for org-managed secret stores. Per-tenant namespace; never returned through the API.
Control-Flow Primitives
Conditional branching, bounded refinement loops, and a Map node for fan-out over collections. Generator-critic patterns are first-class, not client-side hacks.
Knowledge Graph & RAG
Built-in semantic search and entity-aware retrieval via Trailhead. Ground every workflow in your organization's documents and data.
Provider-Agnostic Blob Storage
S3, GCS, Azure Blob, or local — behind one trait. Operators register named storage profiles; workflows reference them by name. Swap providers with zero workflow changes.
Scheduled & Webhook Triggers
The grove-scheduler workspace member runs cron triggers with HA-safe Postgres claims. Inbound webhook endpoints fire a workflow on every POST. Multi-replica safe; retries with exponential backoff.
Autonomous Agent Runtime
Define an agent once — name, system prompt, tool allowlist, model — then run it against goals. Per-run sandboxed workspaces for file/shell/git tools, plus your own external MCP servers from a per-tenant catalog. Sub-agent delegation with depth and fan-out caps. Tenant-scoped git credentials.
Grove as MCP Server
Workflows and skills surface to MCP clients (Claude Desktop, Claude Code, Cursor) as tools and prompts. grove__* provisioning built-ins let an MCP client author workflows and agents over the protocol. Two-key Invoke/Provision pattern keeps the LLM's authoring rights bounded.
First-Class Multi-Tenancy
Real tenant entity in Postgres with row-level isolation across workflows, runs, sessions, agents, skills, and secrets. Per-tenant quotas, rate limits, and budget caps. POST /tenants provisions a tenant and mints its first key in one call.
Sessions & Memory
Multi-turn conversation context with rolling summarization and hierarchical memory namespaces. Tool-based prompt enrichment lets agents promote relevant memories into the system prompt on demand.
What You Can Build
Agentic Research Systems
Automate due diligence, competitive analysis, and compliance research with multi-step AI workflows that query your internal data and deliver structured reports.
Learn more →Document Processing Pipelines
Process contracts, filings, and technical documents at scale. Parallel extraction across financial, legal, and technical dimensions — unified into a single analysis.
Learn more →Multi-Agent Collaboration
Deploy teams of specialist AI agents that analyze different dimensions of a problem in parallel, then merge findings into board-ready deliverables.
Learn more →Autonomous Agents in a Sandbox
Long-horizon AI tasks — investigate a failure, implement a feature, audit a repo — running in a per-run container sandbox with allowlisted file, shell, and git tools. Durable resume; budget caps; sub-agent delegation.
Learn more →LLM-Authored Workflows over MCP
Point Claude Desktop, Claude Code, or Cursor at Grove's MCP server and let the assistant build and publish the workflow it will then invoke. Two-key Invoke/Provision pattern bounds the LLM's authoring rights.
Learn more →LLM Layer for Data Pipelines
Drop a Grove step into Airflow, dbt, or Dagster for classification, extraction, data-quality triage, or generate-and-vet checks. Not an orchestrator replacement — the LLM-heavy step your existing orchestrator was missing.
Learn more →- AES-256-GCM Encryption
- Air-Gap Compatible
- Audit Logging
- Self-Hosted — Your VPC
- Enterprise License Tiers