Tegy runtime change · 13 July 2026
Doctrine is moving to Grok 4.5
Doctrine keeps the same Tegy name and StrategyOS experience, but gains Grok 4.5's vision support and a larger working context. It also becomes the starting model for people who have not chosen one before.
Verified and rollout-readyWhat changes for the team
Before
Doctrine used Kimi K2.7 Code. Playbook was selected for people without a saved model choice.
After
Doctrine uses Grok 4.5. New arrivals begin on Doctrine; anyone with a saved preference keeps it.
Capability and runtime comparison
| Area | Previous Doctrine | New Doctrine | Result |
|---|---|---|---|
| Provider model | Kimi K2.7 Code | Grok 4.5 | Changed |
| Working context | 262,144 tokens | 500,000 tokens | +91% |
| Image input | Enabled in Tegy | Enabled and live-tested | Passed |
| Reasoning | Adaptive, summarized | Adaptive, summarized | Preserved |
| Subagents | Playbook | Playbook | Preserved |
| Output ceiling | Tegy set 32K | Provider-managed; no artificial 32K cap | Improved |
How we tested it
| Check | Evidence | Status |
|---|---|---|
| Real vision turn | Random text was rendered into an image and correctly read by Grok through the actual Tegy agent runtime. | Passed |
| Fair agent conditions | Claude Agent SDK 2.1.165, StrategyOS, runtime skills, 56 available skills, and 38 provider tools. | Matched |
| Endpoint shape | OpenRouter Anthropic Messages route, one image block, adaptive summarized reasoning, no implicit 32K output cap, HTTP 200. | Passed |
| Usage limits | The final provider event reported an exact cost of $0.051944 and entered the normal Tegy usage ledger. | Passed |
| Application regression | Production build, full flow suite, and all 24 composer/model-selector browser tests. | Passed |
Cost watch-out
The small live vision check consumed $0.051944, about 20.8% of the current $0.25 free-session allowance. Limits are intentionally unchanged, and Grok's exact provider cost counts toward them. We should monitor real strategy sessions before changing allowances.
Decision summary
Proceed with Grok 4.5 as the Doctrine provider and Doctrine as the no-preference default. The transition removes Kimi-specific routing instead of keeping a legacy fork. Existing model selections are respected.
Sources: OpenRouter xAI catalog, OpenRouter Anthropic Messages API, and xAI Grok 4.5 documentation.