We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
Architecture: real measured API-vs-CLI-subscription token numbers (section 8) Replace the estimated per-backend token figures with measured ones: the same broad question costs ~340k input tokens over 6 CLI-subscription calls vs ~4k over 4 raw-API calls (~50x/call), and note the raw-API run proved provider portability (a different vendor's model followed the text protocol and passed the eval judge).
Architecture: document the multi-call agent loop (section 8) Add a section on the request-time ReAct loop (POST /chat): what each LLM call does (think / generate / compose / judge), how token cost compounds across sequential calls on API vs subscription backends, and the pros/cons + when-not-to-use-it. Includes a genericized teaching case on quality-gating a token optimisation. Also update two stale 'roadmap' references now that the agent loop is built.
Red-Teaming: normalization pre-pass shipped (#31) — before/after + honest residuals
Red-Teaming: add the concrete measurement-instrument finding (the ~75% was quoted-while-refusing FPs)
Add Red-Teaming page: adversarial injection testing + the oracle-convergence teaching case; link from Home
wiki: teaching pass — upgrade Architecture/Evaluation, add Design-Decisions page
wiki: overview, architecture, evaluation, engineering notes
Initial Home page