Capability statement · July 7, 2026 · Updated October 18, 2026 · 5 min read
Kaptain reduces the operating cost of agentic AI
Kaptain is CHYNJ's model-agnostic Agent Ecosystem OS: local-first, reachable from your own devices over Tailscale, licensed, protected by an access code, and built for cost control, workflow visibility, and scalable agent operations.
Executive brief
AI agents create recurring operating cost. They repeat context-heavy work, overuse premium models, and turn long workflows into unpredictable token bills.
Kaptain is CHYNJ's paid coordination layer for agentic work: local runtime, remote access over your own Tailscale network, model choice, Krew tools, KodeGraph context, licensing, and workflow visibility in one operating environment.
The product thesis is simple: keep frontier-model reasoning available when it changes the outcome, while routine evidence work, tool calls, local execution, and code-context preparation move through cheaper and more controlled routes.
Market pain
Token spend is becoming operating spend. Agentic systems do not answer once; they search, retry, route, summarize, and keep context alive. That changes AI cost from a prompt-level issue into a persistent workflow-level issue.
Benchmark evidence
The benchmark claims below come from two internal reports in the Kaptain repository, listed under Sources. Lower token counts are cited only where recall is comparable.
Fewer Codex tokens
June 25, one exact-lookup task on an earlier build: 11,438 total Codex tokens when Kaptain ran the KodeGraph lookup and handed Codex the result, versus 52,241 for Direct Codex. Gold-file recall was 1.000 in both runs.
Fewer tokens than a standard graph MCP
July 5, three Go-repository tasks, Codex calling Kaptain’s KodeGraph tools: 159,203 average total tokens, versus 322,911 with a standard code-graph MCP and 176,861 for Direct Codex (10.0% fewer). Recall was 1.000 in all three conditions.
Clean June 25 benchmark result
One exact-lookup task, same 1.000 gold-file recall: Direct Codex used 52,241 total tokens; Kaptain + KodeGraph used 11,438. In that run, Kaptain ran the lookup first and handed Codex the result, which also cut uncached input tokens, output tokens, tool calls and wall time.
| Metric | Direct Codex | Kaptain + KodeGraph | Result |
|---|---|---|---|
| Gold-file recall | 1.000 | 1.000 | Same answer quality |
| Total Codex tokens | 52,241 | 11,438 | 78.1% lower |
| Uncached input tokens | 19,444 | 9,377 | 10,067 fewer |
| Output tokens | 413 | 141 | 272 fewer |
| Codex tool calls | 2 | 0 | Kaptain ran the lookup first |
Current Capability and Next Steps
Impact
Kaptain is designed to reduce unnecessary premium-model usage while keeping frontier reasoning available for planning, synthesis, and high-value decisions.
The impact case is operating leverage: lower AI spend, fewer avoidable cloud-model workloads, better use of local compute, stronger workflow control, and support for sensitive workflows that should remain local and private.
Selected benchmark runs show material token reduction: 78.1% on one exact-lookup task (n=1, on an earlier build), and on July 5, across three tasks, 50.7% fewer tokens than a standard graph MCP and 10.0% fewer than Direct Codex. Kaptain aims for this by giving models a code index to query instead of whole files to read, and by letting a cheaper or local worker model take mechanical work.
Support requested
CHYNJ is seeking mentorship, technical feedback, early users, capital support for broader benchmark infrastructure, and introductions to teams experiencing AI cost pressure in agentic development workflows.
The goal is to move Kaptain from working MVP and early measurements into a disciplined launch: stronger benchmark coverage, reliable onboarding, and a credible path from individual builders to team adoption.