News
Releases and the thinking behind them — what changed in Slipstream, and why.
8 September 2026
OpenAI's new frontier model shipped with a 1M-token context window. Slipstream's proxy is provider-agnostic, so it compresses context for Astra traffic the same as any other model — no update required. One honest caveat: exact savings numbers depend on headroom's tokenizer catching up, and you can improve them yourself today in Settings → Models.
Read the article →7 September 2026 · v0.2.2
Every Ubuntu AppImage we'd shipped failed to provision its own Python runtime on first launch — the AppImage launcher's own bundled Python was leaking into our managed processes. v0.2.2 fixes it and we validated the fix on real Ubuntu 22.04 and 24.04 containers before writing this. Also: Headroom 0.37.0, RTK 0.48.0, uv 0.12.10, and a Codex ChatGPT-login fix.
Read the article →17 June 2026 · v0.2.1
Slipstream was adding Anthropic's free prompt-cache discount into "Cost saved" — a number you get whether or not we're running. v0.2.1 takes it out, so the headline counts only the work Slipstream actually does. Your number drops, on purpose, because an honest figure beats an inflated one. The app also updates itself now, starting with this release.
Read the article →16 June 2026
Even the best frontier models still waste the majority of tokens and context on bad architecture. One pattern — domain boundaries + events — turned our AI agents into a true 10x multiplier. It compounds over time and works even better with local open-source models. Here's the data and why it matters for long-term shipping.
Read the article →14 June 2026 · v0.1.5
Slipstream launched compressing the conversation with headroom. v0.1.5 adds a second engine — RTK — that compresses something headroom never touched: the output of the commands your agent runs. The two sit at different points in the pipeline and their savings stack. Here's the difference, why we added it, and how to turn it on — plus an honest note on what we're still learning about the combined effect.
Read the article →