3 ms·
Novgraph: Knowledge graph for codebases that saves you 99% percent tokens
Novgraph is a persistent knowledge graph for codebases .
Some of the features:
Recorded intent.
Ranked blast radius.
Architecture context.
Precise code discovery.
Continuous history. .
Visible context accounting. /novgraph savings
Efficient sessions.
Freshness awareness.
Token savings in range of 96.04% to 99.81%
To try Novgraph:
https://trynovaya.com/
https://github.com/Novaya-AI/novgraph
If any question , I am open to answer questions in the comments
- jahala 15d agoCost per correct answer. cost_per_correct = total_spend / correct_count Why this metric. "-X% tokens removed!" is a marketing number if it ignores whether the answer was right. A tool that saves 90% of tokens but makes 20% more mistakes has worse cost-per-correct. I found that token savings is really not worth looking at. Benchmark a baseline (without novgraph) vs novgraph on a set of tasks. Often with such tools, the agent requires more turns and cache hits are fewer so the actual cost ends up similar, but the agent actually takes more time to complete. Build your own benchmark or try mine: https://github.com/jahala/copeca https://github.com/jahala/copeca .. Haven't touched it for 3 months, so not sure if it's still up to scratch.
- Shourya_55 15d agoAlthough accuracy and token savings are independent measurements Novgraph achieved 85.7% accuracy across seven evaluated tool results, while measurable retrieval responses were 96.04%–99.81% saving tokens.