Project: agentspace.directory is an open-source framework and discovery hub for LLM-agnostic, file-based AI agents. Built on the principle of Single-Source-of-Truth (SSoT), it allows developers to declare an agent's entire "soul"—its identity, logic, memory, and activity—within a single, portable directory of Markdown files.
By treating the file system as the primary orchestration layer, agentspace.directory eliminates fragmented configurations and provides a transparent, version-controlled environment where agents can be developed, tested, and shared regardless of the underlying language model.
The entire agent behavior is declared in one single directory. It can be run seamlessly by different engines:
The website facilitates a "Directory-First" approach to AI. All agent logic is declared in a single, LLM-agnostic folder structure.
instructions.mdThe "DNA" of the agent—system roles and logic.
memories.mdShared long-term context and personal facts.
tools.mdUnified API and capability definitions.
tracks/A subfolder for active state tracking (e.g., debug-xxx.md, feat-xxx.md).
apitests (apitests)Purpose: TDD for agents.
Implementation: gentests are created within individual project repositories.
Gemini Instruction: When working on a task in a tracks/ file, Gemini must first check the project repo for gentests and run them to verify any logic changes.
apistubs (apistubs)Purpose: Environment isolation and safety.
Implementation: Used to mock unstable or expensive third-party APIs.
Gemini Instruction: If a task requires an external API that isn't available in the cloud VM, Gemini should use apistubs to create a mock endpoint based on the schema in tools.md.
Focus: Automated debugging and feature implementation.
Workflow: Reads a bug report in tracks/, identifies the failing apitest in the repo, and fixes the code until the test passes.
Highlight: This workflow heavily leverages the apitests extension (apitests) to practice Test-Driven Development (TDD) for agents, ensuring robust logic changes through automated gentests!
Focus: Business process automation and governance.
Workflow: Uses instructions.md to vet vendors against budget rules. Uses apistubs to simulate a payment gateway response before finalizing a purchase "track."
Focus: Knowledge synthesis and schedule optimization.
Workflow: Declares personal events (e.g., "Kate's Birthday March 19") in memories.md. The agent proactively creates a tracks/ file to plan the event 14 days in advance.