Real token savings on your context, verifiable by you
Trajbl is a token-saving evidence-packing layer between retrieval and the LLM. In this demo the actual Trajbl method runs live on your input and returns a compact evidence packet plus a real, measured token reduction.
It runs as a black box: the Trajbl algorithm stays server-side and protected — you see real inputs and real results, not the method itself. The result is deterministic (same input → same packet, re-run to confirm), and you can check the quality yourself in your own LLM.
Request evaluation access Login
Access is granted manually: request access → we approve you → log in → run an evaluation.
A · Paste your own context
Submit a question and your own plain-text context. Trajbl compresses it live and shows the packet plus the before/after token reduction — on your own material.
B · Prepared English demo corpus
Pick a curated English question over a small sanitized corpus. Trajbl runs real precomputed semantic retrieval, then its compact_v2 / pool12 evidence selection.
What actually happens here
- Real Trajbl, live. The packet is produced by the genuine compact_v2 / pool12 selector — not a mock or a fixed example.
- Real, measured reduction. The token counts are computed from the actual packet, and the result is reproducible.
- You are the judge. Click Copy full verification prompt, paste it into your own LLM (ChatGPT, Claude, Gemini…), and compare the answer from the full context vs from the compressed packet. If they’re close, Trajbl kept what mattered — at a fraction of the tokens.
- Our IP stays protected; please mind yours. The Trajbl method stays server-side. Your submissions and their results are stored for our evaluation review, so please don’t paste confidential data.
Honest scope
This is a deliberately small, self-serve demo — it shows the mechanism on short inputs and does not run an automated quality score. For a realistic evaluation on your own data and scale, or to discuss integration, contact us directly. Please don’t paste confidential, personal, medical, legal or financial data.