Skip to main content

Release Notes

Chronos is distributed as a Go module (github.com/spawn08/chronos), cross-platform CLI binaries, and a container image. Tagged releases include checksums, an SBOM, and signed artifacts. See CLI Install for installation options.

v0.12.2​

  • Release maintenance — publishes the current Chronos main branch as a patch release with refreshed release metadata and artifacts.

v0.12.1​

  • YAML-defined skills — agents can declare skills in YAML config, loaded via a new sdk/skill loader with metadata and versioning support.
  • MCP client/adapter improvements — refinements to the MCP tool adapter and client.
  • deploy CLI cleanup — simplified cli/cmd/deploy.go and expanded sandbox-deploy.yaml example.
  • Team hierarchy/swarm fixes — small correctness fixes in sdk/team.

v0.12.0 — Azure reasoning and observable team streaming​

  • Azure Responses API — native Azure reasoning automatically uses /openai/v1/responses, supports function tools, and preserves encrypted reasoning state across stateless tool rounds.
  • Observable team streaming — team runs honor YAML streaming preferences, forward provider-approved reasoning summaries, expose effective runtime settings, and report stream failures instead of silently falling back to blocking execution.
  • Reliable traces — SQLite and PostgreSQL trace writes now upsert span completion data, including output, errors, and ended_at; CLI output resolves relative SQLite paths so trace databases are easy to locate.
  • Runtime overrides — --no-debug and --no-trace explicitly override enabled YAML settings, complementing the existing positive flags.
  • Safer diagnostics — PostgreSQL DSNs are redacted from configuration output while Azure model deployment names and reasoning-summary settings are displayed clearly.

v0.11.0 — CLI runtime control​

  • Session approval bypass — enter a at an interactive tool prompt to approve the rest of the CLI session, use --permission-mode auto_approve, or use the explicit --dangerously-skip-permissions shortcut. Explicitly denied tools remain blocked.
  • Declarative tool policy — YAML tools now accept permission, requires_confirmation, and requires_user_input; agents accept permission_mode.
  • Streaming preference — YAML stream is honored by REPL and headless runs, with --stream and --no-stream overrides.
  • Native reasoning — normalized reasoning configuration maps to OpenAI reasoning effort, Anthropic thinking budgets, and Gemini thinking configuration; reasoning output remains separate from final answer text.
  • CLI observability — YAML/CLI debug and trace controls wire model/tool spans for blocking and streaming execution.
  • Progressive YAML gallery — the documentation now separates simple agents, intermediate routers/pipelines, advanced teams, production governance, provider recipes, and CLI reference into focused pages.

v0.10.2 — CLI approval reliability​

This patch release improves interactive CLI approval handling for tool calls that require human review.

  • Interactive approvals — the CLI now prompts correctly when an agent requests approval-gated tool execution, keeping terminal sessions responsive and explicit.

v0.10.1 — YAML provider resolution​

This patch release strengthens YAML-based agent configuration.

  • Model provider resolution — YAML model configuration now resolves providers more defensively, improving reliability for declarative agent setups.

v0.10.0 — Production agent platform​

This release focuses on the capabilities required to move from prototype agents to durable, governed, production-ready systems.

Highlights​

  • Deep agent harness — planning, virtual filesystem, context compaction, semantic memory recall, and context-isolated subagents can now be enabled together through harness.NewDeepAgent(...).
  • Durable delegation — subagents run in isolated contexts, can be defined dynamically, and execute through the same durable queue used by long-running agent work.
  • Interoperability — Chronos agents can integrate with MCP, A2A, and AG-UI for tool sharing, agent-to-agent collaboration, and standard UI event streams.
  • Memory and knowledge improvements — semantic recall, batched embedding, chunking, query caching, and relevance thresholds improve retrieval quality for larger workloads.
  • Evaluation workflow — trace capture, dataset generation, evaluator runs, CI gates, and historical trend queries are available through chronos evals commands.
  • Operational foundation — distributed execution, checkpoints, approvals, rate limits, tracing, audit logs, hardened auth, sandboxing, Helm deployment, and signed release artifacts are part of the standard platform.

Key flows​

Documentation​

For commit-level details, review the tagged release on GitHub Releases.