Portable conversational kernel
Meta Semantic Kernel.
The conversation may end. The work should not restart from zero.
A user-owned repository connects the present chat to the sources, state and relationships that must continue. The kernel stays direct when the current context is enough and re-enters persistent memory only when it can materially change the resultant.
Public release 0.3.0 · Apache License 2.0 · 20 August 2026
Kernelrecognises · composes · continues
useful work can leave reviewable continuity
What it brings
Continuity without turning memory into an autopilot.
The kernel preserves enough to recognise the point reached, but it does not confuse what is written with what the host can actually do.
Sources, inferences, state, available capabilities and authority over effects remain distinct. Each connected project continues to own its sources; the kernel fork preserves only the continuity that belongs to the user.
- Re-entry
- Retrieves the relevant working set when the chat is not enough.
- State
- Preserves the current point, pending relations and recovery.
- Operations
- Represents requests, results, receipts and recovery.
- Evolution
- Records reusable improvements without applying them opaquely.
One package, distinct organs
The core remains portable; each host receives its own adapter.
Version 0.3.0 integrates the first complete set without pretending there is an agentic runtime where only a turn-reactive chat exists.
- 01Adapter
Translates the kernel into the host's real capabilities.
- 02Kernel
Orients the present and activates the relevant faculties.
- 03State
Holds continuity, cursors and persistent relationships.
- 04Operations
Separates intent, effect, result and receipt.
- 05Evolution
Preserves reviewable deltas when the work changes the system.
Available adapter
Configuration for ChatGPT.
You need a GitHub account, a copy of the repository under your control and the Custom Instructions space in your ChatGPT account. GitHub connectivity must be available in the host for the chat to read the persistent continuum.
Capabilities depend on the plan, account and tools actually available in the turn. Configuration does not create background processes or grant authority over external effects.
python scripts/configure_chatgpt_adapter.py \
--github-user YOUR_ACCOUNT \
--repository YOUR_FORK- 1Create your fork
User state remains separate from the upstream project.
- 2Generate the configured text
The script inserts your account and repository into the adapter.
- 3Install the Custom Instructions
Copy the complete generated text into your personalised instructions.
- 4Connect GitHub and work normally
The kernel re-enters the repository only when the persistent relationship is needed.
Other cloud providers
The core does not belong to a single service.
The same architecture can receive adapters for other conversational hosts that provide persistent instructions and access to GitHub or an equivalent surface.
Version 0.3.0 implements the ChatGPT adapter. Other providers are an architectural possibility, not a compatibility claim: each adapter must respect the real capabilities of its host.
Relationship with MAIOS
An open-source companion to the AI Setup path.
AI Setup prepares the project and, in the self-configuring route, its Project Kernel for Codex. Meta Semantic Kernel is distinct: it brings continuity to conversational hosts and can be offered as a bonus component. It does not replace the MAIOS package and does not depend on RepoKernel to operate.
Verified state
First integrated incarnation, released as 0.3.0.
The repository structure, configurator and validator are available under the Apache License 2.0. Structural validation passes. Host behaviour, assimilation over time and portability to a second host remain to be observed through real use.
ChatGPT is a product of OpenAI and is named only to identify the available adapter. Meta Semantic Kernel is an independent project and is not supported, certified or endorsed by OpenAI.