How supercompress/supercompress works
SuperCompress is a query-aware LLM context compression product, delivered through a hosted API/Python client and coding-agent integrations. The main workflow sends context plus an unchanged query to a compressor, which returns reduced text and token statistics; optional CCR preserves removed text for retrieval. The graph also shows the local agent proxy path, hosted account/billing capabilities, and framework integrations. README documents capabilities beyond the sampled files; uncertain internal wiring for those is omitted.
Main entry points: compress.js Compress API · engine.js Compression engine · retrieve.js Retrieve API · client.py Python client
Components 22
Product entry points
- Python client
supercompress/client.py - Compression result
supercompress/result.py
Compression runtime
- Compression engine
api/_lib/engine.js - Local engine
packages/proxy/src/local-engine.js - Shared engine asset
web - Compressed context and stats
Hosted platform
- Compress API
api/v1/compress.js - CCR originals
api/_lib/engine.js - Retrieve API
api/retrieve.js - API key management
api/keys.js - Account and usage
api/account.js - Billing endpoints
api/billing.js - Stripe webhook
api/billing-webhook.js - Billing and ledger
api/_lib/billing-ledger.js
Coding agents
- MCP tools
mcp/server.js - LLM proxy
packages/proxy/src/server.js - Proxy compressor
packages/proxy/src/compressor.js
App integrations
- Framework adapters
integrations
Other components
- Developer or agent
- Stripe
- Firebase services
- LLM provider
Connections 27
- Developer or agent → Python client: calls
- Python client → Compress API: posts context/query
- Compress API → Compression engine: compresses
- Compression engine → Shared engine asset: loads engine
- Compression engine → Compressed context and stats: returns text/stats
- Compress API → CCR originals: stores originals
- Developer or agent → MCP tools: invokes tools
- MCP tools → Compression engine: compresses/retrieves
- MCP tools → Retrieve API: retrieves fallback
- Developer or agent → LLM proxy: sends LLM request
- LLM proxy → Proxy compressor: compresses messages
- Proxy compressor → Compress API: requests compression
- LLM proxy → LLM provider: forwards request
- Framework adapters → Compress API: calls compression API
- Python client → Compression result: builds result
- Developer or agent → Retrieve API: requests original
- Retrieve API → CCR originals: reads owned text
- Retrieve API → Firebase services: loads stored record
- Compress API → API key management: authenticates key
- API key management → Firebase services: checks key
- Developer or agent → Account and usage: manages account
- Developer or agent → Billing endpoints: manages plan
- Billing endpoints → Stripe: creates billing actions
- Stripe → Stripe webhook: sends events
- Stripe webhook → Billing and ledger: applies updates
- Billing and ledger → Firebase services: persists ledger
- Account and usage → Firebase services: reads account data