Skip to main content

​
0.2.2

​
Fixed

  • Prompt caching — system prompt is split into a stable prefix and dynamic suffix with Anthropic cache breakpoints so the prefix is reused across turns
  • Rolling cache breakpoint on conversation history (penultimate user message) for multi-turn token savings
  • Read OpenAI cached_tokens from prompt_tokens_details for unified cache reporting

​
Added

  • Per-run ID logging via ContextVar for tracing agent runs in logs
  • Usage line now shows cache write tokens, prefixed with Usage:

​
0.2.1

​
Added

​
0.2.0

​
Added

  • operator init command to scaffold ~/.operator/ with starter config and agent

​
0.1.0

​
Added

  • Initial public beta release
  • LiteLLM-backed agent with model fallback chains
  • Slack transport with Socket Mode and thread continuity
  • SQLite persistence with WAL mode
  • Scheduled jobs with cron, prerun gating, and postrun hooks
  • Vector memory with harvesting and semantic search
  • Key-value store with TTL support
  • 16 built-in tools (shell, file, web, memory, KV, jobs, messaging, sub-agents)
  • Skills discovery from markdown files
  • Turn-safe context truncation
  • CLI for inspection, service management, and job control