Build Intelligent AI Agents
TeamSpark AI Workbench is a powerful desktop and command line client application for Mac, Windows, and Linux. It allows you to easily build intelligent agents for solving hard problems using AI models from providers like Anthropic, OpenAI, Google, AWS Bedrock, and Ollama.
TeamSpark AI Workbench allows you to build sophisticated intelligent agents designed for solving complex problems. These agents are equipped with memory via References, enabling them to retain and utilize information across interactions. They can learn and refine their approach through Rules, which function as dynamic system prompts guiding the AI's behavior and problem-solving strategies. Agents can also utilize Tools to interact with external systems and perform actions, extending their capabilities significantly.
With TeamSpark AI Workbench, you maintain complete control over the precise context sent to the AI, ensuring your agent operates exactly as intended. This powerful combination of features allows you to create highly capable and customized AI agents quickly and easily.
Product Demo Video
Watch an 8-minute speed run, followed by some more complex use case examples:
Download TeamSpark AI Workbench
Get the desktop client for your operating system:
Key Concepts
Providers
TeamSpark AI Workbench supports a wide range of AI model providers, giving you the flexibility to choose the best model for your task.
Chat Session
The chat session is the central workhorse where you interact with AI models, explore ideas, and solve problems. You can select providers/models, override settings, manage context, and review usage.
References
References are how your agent collects and remembers information, serving as the "memory" of a workspace or chat session. They provide background information, guidelines, or specific details that the AI can reference.
References can be any kind of information – text you create, content from documents or web pages, summaries of AI chat sessions, or tool call results. In addition to content, each reference has a name, description, priority level (000-999), can be enabled or disabled, and can specify how it should be included in chat sessions (always, by agent request, or manually). You can manage references manually through the TeamSpark AI Workbench interface.
TeamSpark AI Workbench also provides an internal tool that allows the AI to manage and interact with references. This tool is installed by default and enables the AI to create new references from information it has access to, such as the current chat history or results from other tools. For example, you could instruct the AI to "Summarize the findings from this chat session in a new reference" or "Get the web page at http://somewebsite.com/document and convert its contents into a new reference". The AI can also use the tool to add or remove references from the current chat context when directed by you or indicated by a rule. You can even ask the AI to "Include all references with the word 'cancer' in their description in this chat session."
Rules
Rules govern how your agent instructs the AI to process instructions and commands. They are essentially "system prompts" injected into each chat message to guide the AI's behavior and help the agent learn and apply problem-solving techniques.
Rules are the primary way an agent "learns" and can be used to apply prompt engineering techniques. Note that there is also a System Prompt under Settings for prompts that apply to everything in the workspace all the time. In addition to the rule content, each rule has a name, description, priority level (000-999), can be enabled or disabled, and can specify how it should be included in chat sessions (always, by agent request, or manually). You can manage rules manually through the TeamSpark AI Workbench interface.
TeamSpark AI Workbench also provides an internal tool that allows the AI to manage and interact with rules. This tool is installed by default and enables the AI to create new rules from information it has access to, such as the current chat history or user instructions. For example, you could instruct the AI to "Make a new rule so that you always output properties with attributes in markdown table form" or "Make a new rule for yourself that will ensure that when we encounter a challenge like this in the future you will process it using the technique that was successful above". The AI can also use the tool to add or remove rules from the current chat context when directed by you or indicated by another rule. You can even ask the AI to "Include all rules with the word 'research' in their description in this chat session." or "Review all rules and include any that appear to be relevant to the current chat session." This capability could even be turned into a rule itself, allowing the AI agent to review and include relevant rules automatically in every chat session.
Tools
Tools are specialized functions or capabilities that extend the AI's abilities beyond just conversation, allowing it to interact with external systems, process data, or perform complex operations.
Tools are automatically available to the AI when processing messages and help it accomplish tasks more effectively and provide more comprehensive assistance. They are essential for building intelligent agents that can interact with the real world or specific data sources.
TeamSpark AI Workbench uses the Model Context Protocol (MCP) to manage and interact with tools. MCP is an open protocol that standardizes how applications provide context to LLMs, similar to how USB-C provides a standardized way to connect devices. This open standard ensures flexibility and interoperability.
For more information about MCP, you can visit the official documentation at modelcontextprotocol.io. You can also explore the official collection of MCP servers (implementations of tools) at github.com/modelcontextprotocol/servers.
Workspaces
All activity takes place in a workspace, which is a folder storing your prompts, settings, providers, references, rules, and tools. You can use one workspace or create multiple for different projects or agents.
Settings
Workspace settings, including the system prompt, chat session parameters, and tool configurations, apply to your workspace and can be overridden for specific chat sessions or tools.
CLI Mode
TeamSpark AI Workbench also includes a command line interface (CLI) or "terminal" mode, allowing you to interact with the application and perform tasks directly from the terminal.
This mode is useful for advanced users and automation, and also allows you to run TeamSpark AI Workbench inside of other tools, like an IDE or text editor.
See the READEM.md for details on running in CLI mode.
