ContextForge rebuilds small LLM context for long-running work

helps an LLM handle long tasks without carrying the full forward at every step. The main idea is to treat the as a temporary work area, not as . Each step rebuilds a small context that contains only the information needed for the next action.

A structured LLM Wiki layer was added so knowledge is stored in a persistent, searchable form instead of flat logs. That made longer runs more consistent. The public includes benchmark results for long-running setups, including 180-day and 500-day style tests.

The work is also being compared with RecallBench, which checks more than simple information lookup and is meant to test memory behavior across longer tasks.

Key points

  • rebuilds a small relevant context at each step instead of reusing the full history.
  • The is treated as a work area, not .
  • LLM Wiki stores knowledge in a persistent and searchable structure.
  • Benchmark results include long-running 180-day and 500-day style tests.
  • RecallBench is used as a comparison point for long-task memory behavior.
Read original