Think with context you own.
SelfContext is a local, portable context layer for AI tools. It helps you carry goals, decisions, projects, preferences, and evidence across conversations and tools, so you can continue thinking instead of starting over.
Your durable context lives in a plain-Markdown vault/ that you can inspect,
edit, copy, and back up. SelfContext is not a standalone chatbot or hosted
memory service; your existing AI tool remains the interface.
Normal use needs no server or dependency installation.
git clone https://github.com/joacod/self-context.git
cd self-contextOpen the repository root in an AI tool that loads project-local Agent Skills, then ask it:
ingest my resume into SelfContext
On the first ingest, SelfContext creates vault/ and stores the useful context
there. If you already have a vault, place it at vault/ before using the
repository.
These are natural-language requests to your AI tool, not shell commands:
| Request | Use it when |
|---|---|
ingest ... | You know exactly what fact, document, correction, or decision to save. |
query ... | You want to retrieve context or think through a question using it. |
checkpoint ... | A meaningful conversation may contain durable outcomes, but you want SelfContext to decide what is worth keeping. |
For example:
what does my context say about X?
help me think through X using my context
compare these options against my current goals
challenge this idea based on what you know
checkpoint this discussion
what from this conversation is actually worth keeping?
show me the context behind that recommendation
review my context for stale or conflicting information
Query is read-only by default. Checkpoint does not save a transcript or every brainstorm: it keeps only worthwhile facts, decisions, corrections, or reusable conclusions, and may leave the vault unchanged.
Context can cover cross-domain goals and preferences, as well as:
| Area | Path | What it covers |
|---|---|---|
| Career | career/ | Career evidence and concepts |
| Learning | learning/ | Knowledge states, gaps, corrections, and progression |
| Writing | writing/ | Evidence-backed communication and writing context |
| Relationships | relationships/ | Shared history, commitments, and open loops |
| Media / Taste | media/ | Reactions to cultural works and evolving taste |
| Ventures / Projects | ventures/ | Initiative lifecycle, decisions, commitments, evidence, and outcomes |
Areas are optional and created only when relevant durable context needs them.
vault/ is the durable source of truth. It is Git-ignored and remains ordinary
Markdown, independent of any particular AI tool or viewer. Do not commit or
force-add it to the repository.
After pulling a newer version of SelfContext:
git pullThen ask your AI tool:
upgrade vault latest
This brings an existing vault up to date while preserving its context and history. See the upgrade workflow for details.
- Vision — the problem SelfContext addresses and its design goals.
- Architecture — how the skill, vault, and deterministic tools fit together.
- Full SelfContext skill — operating rules and routing.
- Ingest workflow
- Query workflow
- Checkpoint workflow
- Skill maintenance — guidance for contributors changing the skill.
Run the repository validation from the repository root:
python3 scripts/validate_repo.pyLicensed under the MIT License.