What we're exploring next. These are research directions, not commitments. Priorities shift as we learn what works.
In progress & high priority
Refine Chippery modes for Claude Code via hooks
Our Claude Code implementation is still not robust enough. We want to ship something that mirrors the Chippery agent experience much more securely.
Benchmark harness
A proper framework for running distributed tests and making comparative benchmarks easier to reproduce. We want anyone to be able to verify our results or test on their own codebases.
Retrieval memory engine
A smarter relevancy algorithm for deciding what context to retrieve. The current indexer works well on familiar patterns but struggles with unusual project structures. We're exploring better heuristics.
Exploring
Cached token expense calculations
Some providers offer discounted pricing for cached tokens. We want to surface these savings accurately so you can see the real cost of a session, not just raw token counts.
Token-efficient browser testing
Browser automation tends to be token-heavy because of DOM verbosity and screenshot overhead. We're researching ways to integrate testing workflows without blowing through your budget.
MCP token efficiency framework
As MCP tooling grows, so does the context window cost of tool definitions and responses. We're looking at patterns for keeping MCP integrations lean.