DeepChat 1.1.1 stops loading every tool into context

DeepChat's 1.1.1 release adds two tool modes, Code and Minimal, that change how much of an agent's tool set sits in its context window. The agent used to load every tool definition up front; now it discovers tools on demand and pulls in only what a task needs.

That matters because tool definitions eat context budget. An agent juggling dozens of tools burns tokens on descriptions it never calls, and a long tool list makes the model more likely to pick the wrong one. Scoping by mode also makes CLI tool access permission-gated.

The release bundles other changes too. A Tape Trace Inspector records agent runs, provider requests, tool calls and Skill usage, so a team can check the record rather than the agent's own summary. Skills, previously scattered, now live in one shared library with per-agent enablement and progressive loading, so a team doesn't need to duplicate the same skill definition for every agent. DeepChat also now detects, installs, repairs and recovers Node.js and uv toolchains on its own, closing off a class of failures where an agent works fine until the environment underneath it doesn't.

Managing Node and uv automatically may prevent more failures day to day than the two tool modes do. Missing toolchains cause a lot of agent tooling breakage, and DeepChat is now trying to own that layer instead of leaving it to the user.