Stigmer: An Open Platform for Building AI Agents and Workflows with Zero Infrastructure

Stigmer is an open-source AI agent platform that supports local SQLite development mode and cloud production mode. It adopts the CNCF Serverless Workflow standard and achieves reliable orchestration via Temporal. Developers can use the same set of CLI, SDK, and resource definitions to build and deploy AI workflows, smoothly transitioning from small-scale experiments to production environments without rewriting code.

With the rapid development of LLM capabilities, turning them into reusable, orchestratable, and maintainable production-grade systems has become a core challenge for developers. Existing solutions face a dilemma: fully managed SaaS platforms have limited flexibility and risk of vendor lock-in; open-source projects require self-building complex infrastructure with high operational costs. Stigmer solves this problem with an open-core model: the same codebase, CLI, and SDK support local SQLite development/debugging and cloud production collaboration, enabling smooth transitions from small-scale experiments to production environments.

Stigmer uses a modular microservices architecture, with core components including:

CLI Client

Written in Go, packaged as a single BusyBox-style binary, with a unified command set (run/apply/get/list/delete/server) and consistent interface across local/cloud environments.

stigmer-server

Core gRPC service implemented in Go, responsible for resource management (agents, workflows, etc.), SQLite persistence (local mode), full-text search (FTS5), and Temporal integration.

Agent Runner

Implemented with Python and Temporal, interacts with LLMs like Anthropic/OpenAI/Ollama via Graphton.

Workflow Runner

Implemented in Go, follows the CNCF Serverless Workflow standard, supports complex control flows (conditional branches, loops, parallelism) and long-running workflows (Continue-As-New mode).

Agents

Declarative YAML definition (similar to K8s), including instructions, MCP server access, and model configuration. Capabilities can be extended by combining MCP servers.

Workflows

Multi-step automated processes that support tasks like HTTP calls and agent invocations. Adopts the CNCF Serverless Workflow standard for ecosystem compatibility.

Skills

Versioned knowledge artifacts, including SKILL.md (YAML metadata) and optional tool scripts/documents, to promote knowledge reuse and sharing.

MCP Server

Based on the Model Context Protocol open standard, provides tool access capabilities, supports STDIO-based servers (npm/Python/Go/Docker), and can integrate with AI-driven IDEs.

Feature	Local Mode (Open Source)	Cloud Mode (Stigmer Cloud)
Startup Method	stigmer server	stigmer backend set cloud
Storage	SQLite (~/.stigmer/stigmer.db)	Distributed managed storage
Orchestration	Temporal (auto-managed)	Temporal (managed)
User Management	Single implicit user	Organizations, teams, IAM
LLM Support	Anthropic, OpenAI, Ollama	Configurable
Use Cases	Development, personal projects, offline environments	Team collaboration, production environments, governance needs

Both modes use the exact same resource definition format and can be switched seamlessly.

• Single Binary Distribution: All components are packaged in BusyBox mode, simplifying installation (Homebrew/shell/source compilation).
• Temporal Integration: Provides reliable state persistence, automatic failure recovery, long-running workflow support, and horizontal scaling.
• Pure Go SQLite Implementation: Uses modernc.org/sqlite, no CGO dependencies, runs consistently across platforms.
• Adoption of Open Standards: CNCF Serverless Workflow (workflows), MCP (tool integration), gRPC (service communication), protobuf (API contracts) — reduces lock-in risk and improves interoperability.

Stigmer is suitable for various AI engineering scenarios:

1. Intelligent Customer Service Automation: End-to-end workflow (understand the problem → query knowledge base → call API → generate response).
2. Code Review Assistant: Analyze code changes, check best practices, identify security issues, integrate with CI/CD.
3. Data Processing Pipeline: Orchestrate data extraction, transformation, and analysis processes; use LLMs to parse unstructured data.
4. Multi-agent Collaboration System: Coordinate collaboration among professional agents for research, writing, review, etc.
5. Personal Automation Assistant: Personal workflows for handling emails, schedules, and task management.

Stigmer provides a practical and elegant solution for AI engineering, combining mature technologies like Temporal, SQLite, and CNCF standards, and offering a smooth developer experience through abstraction layers. Its 'zero infrastructure' promise allows developers to launch the platform in minutes without worrying about deploying databases, message queues, or orchestration engines; the cloud mode provides enterprise-level scalability and governance capabilities. For developers and teams looking to turn LLMs into production-grade systems, Stigmer is a choice worth considering.