Context Vault: project context
Keep a project’s documents, conventions, decisions and constraints in one place and hand a new AI agent or developer a context pack for a specific task — without retelling the history by hand. Files are read in your browser; their contents are sent nowhere.
Free, no sign-up
Your files never leave the browser
The folder and files are read here, in your browser: their contents are not uploaded to an InterLAB server or anywhere else. Source texts are kept in this browser’s IndexedDB; decisions and constraints in localStorage.
Add text
A document, a team convention, a glossary or a task log — paste the whole text.
Files and the project folder
Files .md, .txt, .json, .yaml, .toml, .csv and source code. When a folder is opened, node_modules, .git, dist, build, .next, vendor, coverage, .env files, lock files, binaries and files over 300 KB are skipped; a simple root .gitignore is respected.
Sources
Rules v1.0, 2026-10-13. Checks are hints with evidence; a human decides what to do.
How Context Vault works
Sources
Paste text, choose files or open the project folder. Documents, config and code up to 300 KB per file are read; service folders (node_modules, .git, dist, build, .next, .nuxt, .svelte-kit and more), .env files and lock files are skipped, and a simple .gitignore is respected. The limit is 400 files and 8 MB per folder.
Decisions and constraints
Decisions numbered DEC-001…: context, decision, reason, status (active, superseded, deprecated), date and review date, linked sources. Constraints CON-001…: hard ones go into every pack, soft ones when they match the task. Superseding links both decisions both ways.
Search and staleness
Search the whole vault with highlighted matches and filters. Staleness rules show overdue decisions, files not changed for a long time, references to missing paths and duplicates — with evidence.
Context pack
Describe the task — the tool selects matching decisions, constraints and file excerpts, shows each item’s score and matched words, and fits the pack into a token budget. The pack is Markdown in an AGENTS.md/CLAUDE.md structure plus JSON; it is scanned for secrets before copying.
Rules and method
The rules table is generated from the same code that runs the checks — rules v1.0 (2026-10-13).
CV-S01 Decision review date has passed
- Severity
- medium
- What is checked
- An active decision whose review-by date is earlier than today.
- What to do
- Confirm the decision and set a new date, or supersede it.
CV-S02 Superseded decision in the pack
- Severity
- high
- What is checked
- A superseded or deprecated decision is manually included in the current context pack.
- What to do
- Exclude it from the pack or include the active decision that replaced it.
CV-S03 File not changed for a long time
- Severity
- low
- What is checked
- A file older than N days relative to the newest file in the vault (or today — configurable).
- What to do
- Check that the document still describes the current state; update or remove it.
CV-S04 Reference to a path that is not in the vault
- Severity
- low
- What is checked
- A document mentions a path like src/foo.ts, but no file with that path is in the vault (best effort: documents only, when the vault has files with paths).
- What to do
- The file was renamed, deleted or not loaded — fix the document or add the file.
CV-S05 Duplicate source
- Severity
- low
- What is checked
- Two sources with identical content (same hash).
- What to do
- Keep one — duplicates waste the pack budget.
CV-S06 Decision without a rationale
- Severity
- medium
- What is checked
- An active decision has no “why”.
- What to do
- Write down the reason: without it an agent or a new developer cannot tell when the decision may change.
CV-S07 Linked source was removed
- Severity
- low
- What is checked
- A decision links to a source that is no longer in the vault.
- What to do
- Link the current source or remove the link.
CV-S08 Broken “superseded by” link
- Severity
- medium
- What is checked
- A decision is marked superseded, but the replacement is missing, does not exist, or is itself superseded (the chain does not end in an active one).
- What to do
- Point it to the active decision that replaced it.
CV-S09 No review date
- Severity
- low
- What is checked
- An active decision has no review-by date — rule CV-S01 can never fire for it.
- What to do
- Set a review date (for example, six months ahead).
- Search: code identifiers are split (useToolStore → use tool store, snake_case, paths), lowercase, ё → е, RU/EN stop words; words of 5+ letters lose one common ending (bookings → booking). BM25 ranking with k1 = 1.2, b = 0.75; title and path count 2 times.
- Pack selection: hard constraints always; soft constraints and active decisions when they match the task words; conventions and glossaries when they match; other files — up to 10 best with a score of at least 30% of the top one. Duplicates (CV-S05) and stale files (CV-S03) are never auto-selected — only by hand. Superseded decisions that match the task go to “What not to do”.
- Budget (8,000 tokens by default): the task, constraints and decisions are never cut; files go in order (pinned, then by score) — whole if they fit, otherwise as an excerpt of matching lines ± 3 lines, otherwise as a path only.
- The default file staleness threshold is 180 days from the newest file in the vault (or from today).
Limitations
- Search works on words, not meaning: it does not find synonyms or paraphrases. Add tags and clear titles.
- A file’s modification date is what the browser reports; after git clone or unpacking an archive it reflects the copy time, not the last change in history.
- The path check (CV-S04) is best effort: it looks for mentions like src/foo.ts in documents; imports in code are not analysed.
- .gitignore support is simple: negations (!pattern) and nested .gitignore files are ignored — what was skipped is shown in the folder report.
- The secret scan catches common key formats and assignments like password=…; finding nothing does not prove there are no secrets.
- The vault lives in one browser: to move it to another computer, download the JSON and import it there. There is no collaboration or access control.
Privacy
- Files and the folder are read in your browser. Their contents are not uploaded to an InterLAB server or third-party services; there is no AI model inside the tool — published rules do the work.
- Source texts are kept in this browser’s IndexedDB; decisions, constraints and versions in localStorage. “Clear all” deletes them.
- Analytics receives only the fact of an export and its format — no texts, paths or titles.
- The result reaches us only if you send a request yourself and keep the “Attach the result” box ticked — the form shows exactly what will be sent. The summary holds counts, decision IDs, source paths, the pack task and its size — never source contents or decision texts.
Related tools
All tools · 40
Code reviewer, incl. AI-written codePaste a pull request diff — get findings on security, migrations, tests and architecture with evidence from the code, sorted by severity. The code is never changed.
MVP health check16 questions about code, access, data, tests and releases. Shows what to check first — no “score out of 100”, no sign-up.
Spec Drift DetectorRequirements ↔ Git/PR changes: a traceability map, likely drift, changes without requirements and requirements without test evidence — with a reviewer workflow.
Project handoffRepository snapshot, architecture, progress, blockers, risks, tests and next tasks → a verifiable handoff package for a developer or an AI agent.
Agent cost trackerImport API events (CSV/JSONL) → tokens and spend by project, model and task, retries, budgets and alerts. Prices come from your own price list.
Release readinessAn evidence-backed check matrix: CI, migrations, backups, a rollback plan, monitoring, risk exceptions and sign-off → a GO / NO-GO decision.Want your team to work well with AI agents?
Technical mentorship: how to keep decisions, context and reviews so that agents and people follow the same rules. The first short conversation about the format is free; the cost is set after a brief and an agreed scope.
- The first short conversation is free — we choose the format
- NDA — yours or ours
- Pricing after a brief and an agreed scope