How a Post‑00 Team Open‑Sourced OpenChronicle After OpenAI’s $100/Month Feature

On 2023-04-20 OpenAI released Chronicle, enabling AI agents to see the user’s screen and retain contextual information. Access is limited to ChatGPT Pro subscribers at $100 per month.

48 hours later the developer team “Vida” published the open‑source project OpenChronicle ( https://github.com/Einsia/OpenChronicle), extracting the “eyes and memory” layer as reusable infrastructure.

OpenAI’s Chronicle points to an important future, but AI memory should not be locked behind a $100‑per‑month paywall. So we open‑sourced it.

Key differences from OpenAI’s implementation

Runs entirely locally.

Compatible with any model, including self‑hosted models.

Memory can be shared by multiple AI agents simultaneously.

Community response

Within nine hours the repository generated more than 2 000 discussion posts, and many developers began reproducing and extending the project.

Concrete use cases

Context‑aware pronoun resolution – When the user asks “what’s the bug of that?” without prior explanation, OpenChronicle retrieves the current screen context (e.g., the file open in VS Code and the error message) and resolves “that” to the specific code, producing a correct answer.

Cross‑session continuity – In a test with Claude, the model was asked to write an OpenChronicle logo prompt without any previous mention of the project. Without memory Claude first asks “What is OpenChronicle?”; with OpenChronicle it automatically extracts project information from recent activity in the browser, Feishu and VS Code and returns the prompt directly.

Habit‑aware agents – The system learns that the user uses Google Calendar for work and Apple/Fantastical for personal events. When the user says “Add dinner with my parents this Sunday,” the agent routes the event to the personal calendar instead of the work calendar.

Technical architecture

Memory is persisted as plain‑text Markdown files.

Metadata and full‑text search are indexed with SQLite.

The UI hierarchy is exposed through an Accessibility (AX) Tree, making the data readable, editable and migratable.

Local‑first design allows users to summarise memory with on‑device models, keep all data on the device, and pause or resume recording at any time.

Integration flexibility

OpenChronicle does not bind to a specific model or tool; Claude Code, Codex, OpenCode, Claude Desktop and other agents can be connected with a single click, and the required MCP configuration is generated automatically.

Broader implications

The project separates the “eyes and memory” capability from a product‑centric subscription model, raising three questions:

Who controls the memory – the platform or the user?

Are memory boundaries locked inside individual applications or can they flow across tools?

Is memory a black‑box feature or an open data layer?

By making memory a reusable infrastructure layer rather than a proprietary feature, AI interaction shifts from isolated single‑turn Q&A to a continuous, process‑aware partnership where the agent observes the environment, remembers ongoing work, and participates in workflows.