v1.10.10 v1.0.0
Gentleman-Programming/engramMar 29, 2026 → Mar 30, 20261 day between releases
What's New
Rebrand to Council of High Intelligence adds multi-provider routing but regresses versioning from 1.10.10 to 1.0.0.
Breaking Changes
- Version numbering moves backwards (1.10.10 → 1.0.0), so semver-based pinning, update checks, and package managers will treat this as a downgrade rather than an upgrade.
- The deliberation protocol is restructured from 9 steps to 7, so any custom prompts, wrappers, or docs keyed to the old step sequence will no longer line up.
- Verdict output sections are reordered to lead with Unresolved Questions and Next Steps, which can break downstream parsing or tooling that assumes the previous section order.
- Council member metadata is consolidated into agent YAML frontmatter, so locally customized or forked member definitions may need to be migrated.
- Multi-provider auto-routing is enabled by default with zero configuration, changing default execution behavior unless --no-auto-route is passed.
New Features
- Automatic detection and routing of council members across Anthropic Claude, OpenAI/Codex, Google Gemini, and local Ollama models for genuine model diversity.
- A roster of 18 historical-thinker council members, each with a defined domain and default model.
- Three deliberation modes (Full, Quick, Duo) to scale the depth of a session to the problem.
- A Problem Restate Gate requiring each member to restate and reframe the question before analysis begins.
- New --no-auto-route and --dry-route flags for inspecting or disabling provider routing.
Improvements
- Protocol enforcement collapsed from three separate steps into a single post-round scan, shortening the flow from 9 steps to 7.
- Verdict templates now surface actionable Unresolved Questions and Recommended Next Steps ahead of epistemic scorecards.
- Member metadata unified into agent YAML frontmatter, giving one source of truth per member.
- Outcome tracking added via a Follow-Up section in all verdict templates.
- Routing algorithm separates polarity pairs across providers, spreads members evenly, and honors per-member affinity hints with fallback to Claude on dispatch failure.
Upgrade Notes
Treat this as a product-line change rather than a routine bump: the version number drops from 1.10.10 to 1.0.0, so pin explicitly by tag or commit and verify your tooling does not reject the apparent downgrade. Audit any scripts that parse verdict output or reference the old 9-step protocol, and pass --no-auto-route if you need to preserve prior single-provider behavior. The go install version-resolution fix from 1.10.10 is not mentioned in the newer notes, so confirm version reporting still works after switching.
The Two Releases
v1.10.10
Mar 29, 2026· 7 months ago
Bug fix release ensuring the correct version is reported when installing via `go install` instead of 'dev'.
Full release notesv1.0.0 — Multi-Provider Council
Mar 30, 2026· 7 months ago
The initial stable release featuring 18 historical thinkers, multi-provider auto-routing across Anthropic, OpenAI, Google, and Ollama, and a streamlined 7-step protocol.
Full release notes