AI tools

Use CacheSphere as a small grounding layer for agents and internal tools. Fetch JSON once, cache it locally, and link humans back to detail pages.

Agent entrypoints

• GET /api/agent-entrypoints.json — recommended machine entrypoints.
• GET /api/context-pack-index.json — pack index for selecting the smallest useful build memory.
• GET /api/context-pack-schema.json — schema for Context Packs.
• GET /api/decision-kit-schema.json — schema for shareable Decision Briefs.

Flow: fetch the entrypoint manifest, select packs from the index, then fetch only the packs that match your task.

Primary dataset

GET /api/languages.json returns 134 language entries with descriptions, use cases, snippets, tips, and docs_url links.

Context Packs

Context Packs are the agent-facing shortcut when a task needs implementation memory, not the full catalogue.

• Filter by languageIds, stackIds, projectTypes, requirements, and tags.
• Prefer the smallest pack that still carries the decision context the agent needs.
• Use the pack's examples, idioms, and gotchas to keep the prompt concrete.

Other JSON files

• GET /api/ai-tools.json — 12 AI/dev tool entries with tags and official/docs URLs.
• GET /api/stacks.json — 5 decision stack records for practical paths.
• GET /api/same-task-corpus/index.json — same-task corpus for implementation prompts and comparison.
• GET /api/same-task-snippets.json — compatibility view for the older shared-snippet surface.

Optional API server endpoints

If the Node API server is running, these routes are available. Static mirrors should use the JSON above.

• GET /v1/languages/{id} — one language by ID or name.
• GET /v1/transfer?concept=X&from=A&to=B — side-by-side code and difficulty delta for a shared concept.
• POST /api/recommend-language — advisor endpoint when the API server is available.

Decision Brief shape

The Decision Brief schema is for a compact answer copied into an ADR, issue, or agent handoff: recommendation, alternatives, warnings, evidence links, and Markdown.

Benchmark harness

The first token-cost pilot in /benchmarks/ publishes the plan, fixtures, and result schema. It does not publish measured token-savings results yet.

Useful fields

• id — stable key, e.g. python, javascript, rust.
• overview / description — short human-readable summary.
• common_use_cases / commonUses — typical use cases.
• compareData — snippets for loops, functions, classes, strings, imports, and errors.
• docs_url — relative link to the human detail page.

Integration sketch

const entrypoints = await fetch('/api/agent-entrypoints.json').then(r => r.json());
const index = await fetch(entrypoints.entrypoints.contextPackIndex).then(r => r.json());
const compactPack = await fetch('/api/context-packs/agent-memory-compact.json').then(r => r.json());

return {
  packId: compactPack.id,
  title: compactPack.title,
  summary: compactPack.summary,
  nextSteps: compactPack.agentUse.whenToUse
};

Good use

• Pressure-test whether a recommendation actually fits.
• Explain tradeoffs with a source URL.
• Give agents compact context instead of scraped HTML.
• Return a shareable Decision Brief instead of a vague ranking.

Attribution

CacheSphere is MIT licensed. If a tool quotes or summarizes the catalogue in user-facing output, cite CacheSphere and link to docs_url or the relevant deployed page.

← Back to home