Skip to main content
We believe that an agent can learn from its past, just like humans, and become more intelligent. However, memory mechanisms based on the human brain are very vague. In contrast, humans have a more concrete way of accumulating experience: through note-taking systems. Acontext draws inspiration from Notion’s design philosophy, providing the Experience Agent with a similar environment where it can explore and store its own experiences.

​
Enabling Self-Learning

space = client.spaces.create()
session = client.sessions.create(space_id=space.id)
You only need to connect the Session to a Space when creating it, and Acontext will automatically learn tasks as SOPs.

​
Experience Agent’s Tasks

  • Determine whether tasks are difficult. Acontext is not a conversational memory solution, but rather collects difficult tasks and summarizes them into SOPs.
  • Properly place these SOPs into the Space
  • Search for relevant experiences from the Space

​
WorkSpace Structure

Agent experience will be stored in a structured workspace, with folders and pages:
Structure Example
/
└── github_ops/ (folder)
    └── GTM_sop (page)
        ├── find_trending_repos (sop block)
        └── find_contributor_emails (sop block)
    └── basic_ops (page)
        ├── create_repo (sop block)
        └── delete_repo (sop block)

​
SOP with Conditions

How do we store SOPs? Acontext automatically learns from context, determines the scenarios where this SOP should be used, and collects complete tool-call patterns:
Example
{
    "use_when": "star a repo on github.com",
    "preferences": "use personal account. star but not fork",
    "tool_sops": [
        {"tool_name": "goto", "action": "goto the user given github repo url"},
        {"tool_name": "click", "action": "find login button if any, and start to login first"},
        ...
    ]
}
For the detailed ideas of Acontext, check this blog: Context Data Platform for Self-learning AI Agents.