Distillr v0.1.0

Roadmap

Phase 0 validated the core claim: on five realistic payload types Distillr removes 95% of tokens with 100% needle recall (see benchmarks/RESULTS.md). The phases below follow the specification's order. Each item is a GitHub issue; help wanted ones are open to anyone and good first issue marks small, well-scoped starts. Comment on an issue to claim it.

Milestones: Phase 1 · OSS release · Phase 2 · Hosted wedge · Phase 3 · Expand

Phase 1 · OSS release

LLMLingua-2 in the pipeline, audit polish, self-hosted proxy, PyPI release, positioning. Target: 2026-10-15.

# Task Open to
#1 Wire LLMLingua-2 (Stage 2) into the default pipeline behind the extra help wanted
#2 Self-hosted OpenAI-compatible proxy (FastAPI) + Docker image help wanted
#3 Publish to PyPI with trusted publishing on tag good first issue
#4 Embedding-based ranking for Stage 1 help wanted
#5 Audit checker v2: LLM-judge mode and better token matching maintainer
#6 Benchmark on real public datasets, not only generated payloads help wanted
#7 TOON: match the upstream spec test suite and add YAML output help wanted, good first issue

Phase 2 · Hosted wedge

Ledger on Postgres, dashboard, hosted proxy, usage billing, design partners. Target: 2026-12-15.

# Task Open to
#8 Ledger on PostgreSQL with the same schema; multi-tenant keys maintainer
#9 Dashboard: savings over time, per endpoint, audit-risk trend help wanted
#10 Hosted proxy: API keys, per-org quotas, usage-based billing (Stripe) maintainer

Phase 3 · Expand

Routing, semantic caching, TypeScript SDK, enterprise. Target: 2027-03-31.

# Task Open to
#11 Semantic caching of near-identical requests maintainer
#12 Multi-provider routing: compress once, send to the cheapest capable model maintainer
#13 TypeScript SDK (@distillr/node) help wanted
#14 Enterprise: SSO, audit-log retention, on-prem hosted option maintainer

Principles

Not planned