v0.9.0-alpha — multi-vendor ready · ~94/100 parity

🐺 Okami Agent v0.9.0-alpha — a big capability jump. ~94/100 functional parity with the state of the art (NousResearch/hermes-agent), including multi-vendor readiness. From ~1.7k → 2,408 tests passing.

Public alpha: the command/config surface may still change between alphas. Before exposing it publicly, run okami policy check --strict.

Highlights

• Multi-vendor ready — native Gemini (generateContent) and Bedrock (Converse/IAM) transports. Subscription-Claude today, ready to switch vendors when you need it; SDKs install on demand via okami deps.
• Hardened security — scope-aware threat-pattern library (injection / C2 / anti-forensic / invisible-unicode), context-file injection scan (AGENTS.md/.cursorrules), MCP exfil scanner + OSV malware check, SSL CA-bundle preflight, and Tirith pre-exec content scanning (homograph / pipe-to-interpreter).
• Provider & local-model resilience — reactive error recovery (401-refresh, image-shrink), multi-pass tool-call JSON repair, llama.cpp schema sanitization, stall-vs-truncation.
• Platform UX — tiered display-config, long-turn heartbeat, panic-hook crash log, multi-marker silence, photo-album merge, image auto-extract from text, TTS.
• Automation & extensibility — Blueprints (okami blueprint), Kanban swarm (okami swarm), plugin discovery (okami plugins), CDP browser supervisor, lightweight zero-dep web dashboard (okami gui).

New commands

okami deps · okami blueprint · okami swarm · okami plugins · okami gui · okami completion · okami mcp --auth · okami logs --level/--component/--since · okami doctor --fix

Fixed

~36 real defects found by adversarial subagent bug-hunts and fixed with TDD across research rounds #7–#13.

Full notes: CHANGELOG.md · parity scorecard: docs/COMPETITIVE_RESEARCH_13.md

🤖 Generated with Claude Code