# TELOS

> Context is yours · Agents are hired. A cache-aware inference infrastructure for AI agents — one canonical IR across Anthropic, OpenAI, DeepSeek, vLLM and SGLang.

## Docs

- [SWE-bench Verified A/B](https://docs.telosai.pro/en/benchmark/swebench.md): A pre-registered dual-arm study: TELOS halves the input-token bill at the same correctness band — ratios can be faked, dollars can't.
- [Three-layer Architecture](https://docs.telosai.pro/en/concepts/architecture.md): Harness → Bridge → Engine. The IR in the middle is the only contract; the three layers only pass values downward.
- [Three-color Bands](https://docs.telosai.pro/en/concepts/bands.md): PIN / FOLD / DROP — the clean cut that every content block must land in, and the ordering invariant that wins the cache.
- [The TELOS Protocol](https://docs.telosai.pro/en/concepts/protocol.md): Not compression, but never breaking the prefix. The one structural rule that wins the KV cache.
- [ref-pool](https://docs.telosai.pro/en/concepts/refpool.md): The pointer table for large content — slugs frozen on registration, folded without ever touching the prefix bytes.
- [RTK Output Filtering](https://docs.telosai.pro/en/concepts/rtk.md): An orthogonal layer that shrinks the repetitive tool_result tail — independent of TELOS prefix caching.
- [Harness Integration](https://docs.telosai.pro/en/guides/harnesses.md): Claude Code, Codex, OpenClaw, and Hermes — how each connects, and how TELOS detects which harness a request belongs to.
- [Choosing an Integration Path](https://docs.telosai.pro/en/guides/integration-paths.md): Two functionally equivalent ways to route traffic through TELOS — the out-of-process proxy, or the in-process SDK transport.
- [Multi-turn State Accumulation](https://docs.telosai.pro/en/guides/multi-turn-state.md): How TELOS carries the ref-pool and refresh counters across turns — and how to observe the accumulation.
- [Path B · Proxy Gateway](https://docs.telosai.pro/en/guides/proxy-gateway.md): The zero-intrusion HTTP reverse proxy — start, stop, configure, and how harness auto-detection works.
- [Path A · SDK Transport](https://docs.telosai.pro/en/guides/sdk-transport.md): In-process integration — swap your Anthropic or OpenAI client for the TELOS transport; every .create() call stays the same.
- [TELOS](https://docs.telosai.pro/en/index.md): Context is yours · Agents are hired. No rewrite. No compression. 90% token billing saving.
- [Developer Page](https://docs.telosai.pro/en/observability/developer-page.md): The live in-memory diagnostic view — IR structure, PIN/FOLD/DROP distribution, cache breakpoint slots, and per-tool stats.
- [Replay & Comparison](https://docs.telosai.pro/en/observability/replay.md): Record real sessions and replay them under different modes to measure savings — controlled, cheap, and trajectory-stable.
- [Savings Dashboard](https://docs.telosai.pro/en/observability/savings-dashboard.md): The user-facing dashboard — how many tokens and how many dollars TELOS saved, pinned to absolute cost.
- [About & Citation](https://docs.telosai.pro/en/reference/about.md): Who builds TELOS, the license, and how to cite the protocol and the SWE-bench study.
- [Architecture Reference](https://docs.telosai.pro/en/reference/architecture-reference.md): The authoritative deep dive — core data structures, the bridge primitives, engine strategies, invariants, and extension points.
- [Changelog](https://docs.telosai.pro/en/reference/changelog.md): User-visible changes to telos-sdk. Format follows Keep a Changelog; dates are absolute.
- [CLI Reference](https://docs.telosai.pro/en/reference/cli.md): Every telos subcommand and flag — gateway management, integration, mode switching, dashboard, and replay.
- [Support Matrix](https://docs.telosai.pro/en/reference/support-matrix.md): Harnesses, frontier models, and self-hosted inference frameworks that TELOS supports today.
- [Installation](https://docs.telosai.pro/en/start/installation.md): Install TELOS from PyPI, the one-line script, or source — plus requirements and one-command integration.
- [Quickstart](https://docs.telosai.pro/en/start/quickstart.md): Install TELOS, connect your harnesses, and watch the savings — in three commands.