Changelog

v1.0 — live (May 2026)

Modules — production

• mcp-optimizer — embed tool descriptions, drop irrelevant ones. Local benchmark suite reports 33.4% average tool-token reduction across synthetic 120-tool fixtures.
• exact-cache — hash-based identical-request cache.
• semantic-cache — embedding-similarity cache.
• cost-guard — per-request, per-day, per-month USD caps.
• patterns — persistent memory; injects relevant patterns, forges new ones from outcomes.
• ipc — Infinite Persistent Context; compresses old turns instead of dropping them.
• router — fallback chains + configured model/provider selection.
• rehydrator — pull archived context back when referenced.
• compaction-bridge — survive Claude Code auto-compaction.
• prompt-optimizer — restructure prompts for max provider cache hits.
• guardrails — PII redaction, profanity blocks, custom regex.

Modules — preview

• tool-cache — observes tool calls and reports would-hit metrics; active injection is planned.
• mpp-gate — HTTP 402 discovery, challenge flow, sessions, production settlement, and Payment-Receipt headers for managed MPP.

Storage

• Cloud: managed account-scoped storage.
• Local: local data volume.
• Same StorageAdapter interface across both.

API surface

• POST /v1/messages — Anthropic Messages-compatible.
• POST /v1/chat/completions — OpenAI Chat Completions-compatible.
• POST /v1/agent/messages — MPP agent payment route. Discovery, 402 challenges, paid retries, settlement, and receipts are live for managed MPP.
• GET /v1/pipeline — inspect the active module pipeline.
• /v1/keys — API key CRUD.
• /v1/billing/* — Stripe Checkout, Customer Portal, balance lookup.
• GET /health — liveness + provider config view.

SDK

• @prxy/module-sdk — Module interface, StorageAdapter interface, helper types.
• @prxy/shared-types — canonical request/response types.

Local mode

• Single Docker image: prxymonster/local.
• Free, forever.
• Bypassed auth, no billing, no telemetry.

Auth

• Hashed API keys.

• Short-lived hot-key cache.

• Email/password auth.

• Hosted checkout + customer portal.

• Per-tier monthly quota + auto-recharge.

• BYOK pricing model.

Planned

New modules

• airgap — enforce zero non-provider outbound network calls.

Other

• /v1/export and /v1/import for migrating between local and cloud.
• npm-package module loading (load custom modules without a local file path).
• Open-source dashboard package for local mode.
• Post-hook support for streaming responses.
• Lua INCR-based rate limiter (replaces naive get/set).

v1.2 — planned

• evals module — A/B test prompts and modules.
• replay module — replay historical traffic against new pipelines.
• branch module — fork conversations, compare outcomes.
• Hybrid local-cloud sync.
• pattern-sync module — sync patterns across devices.

v2.0 — long-term

• Q-learning router — learn which model best handles which query type.
• collective module — opt-in anonymized cross-user pattern sharing.
• Multi-region cloud deployments.
• SOC 2 + HIPAA compliance certifications.