Skip to main content
TELOS is adapter-driven: the same IR runs across every target below. Need another harness or model backend? Keep the IR and add an engine/harness adapter without rewriting your agent logic.

Harness support

HarnessTypical usagetelos init auto-connectStatus
Claude CodeAnthropic-native coding agent workflow🟢 First-class
OpenClawOpen-source agent runtime with TELOS parser integration🟢 First-class
HermesMulti-agent orchestration with independent sub-IR handling🟢 First-class
CodexOpenAI-style coding workflow via local gateway injection🟢 Supported
See Harness integration for how each connects and is detected.

Frontier model support

Model familyProviderThrough TELOS engine adapterNotes
Claude (4.x / 4.6+)AnthropicExplicit breakpoints and prewarm path
GPT (4+ / 5.x)OpenAIUses prompt_cache_key routing strategy
DeepSeek (V3+)DeepSeekDeterministic byte-stable prefix behavior

Inference framework support

FrameworkDeployment styleThrough TELOSCache-aware capabilities
vLLMSelf-hosted OpenAI-compatible servingExplicit anchors, prewarm, cache probe/evict, partial fork-and-replace
SGLangSelf-hosted high-throughput servingExplicit anchors, prewarm, cache probe/evict, full fork-and-replace

Engine capability matrix

FieldAnthropicOpenAIDeepSeekvLLMSGLang
explicit_breakpoints
max_breakpoints40022
ttl_controlpresets (5m/1h)presets (in-memory/24h)nonenonenone
prewarmable✓ (max_tokens:0)✓ (max_tokens:1)✓ (prewarm_only)
routing_key✓ (prompt_cache_key)✓ (cache_salt)✓ (affinity_key)
cache_probe
span_eviction
fork_and_replace
tier_hint
Bidirectional classNoNoNoYesYes

Architecture reference

Each engine’s emit strategy and usage-parsing rules.