We built ContextGraph because coding agents still lose too much working state when context gets tight. Most tools either store raw text or collapse a long session into one lossy summary. ContextGraph takes a different path: it checkpoints durable working state like decisions, constraints, open tasks, failures, changed files, and restoration instructions. This release adds three pieces we think matter a lot for real coding workflows: • Reactive Delta Compaction • Branch-Aware Context Cache • repo-local .contextgraph memory directories That means an agent can compact, resume later, branch from a checkpoint, and keep the important state visible inside the repo for both humans and other agents. Would love feedback from anyone building real coding-agent workflows, especially around compaction, handoffs, and branch-based execution.

Forum Threads

p/contextgraph

•

7mo ago

You do not have to lose too much working state when context gets tight.

We built ContextGraph because coding agents still lose too much working state when context gets tight.

Most tools either store raw text or collapse a long session into one lossy summary. ContextGraph takes a different path: it checkpoints durable working state like decisions, constraints, open tasks, failures, changed files, and restoration instructions.

This release adds three pieces we think matter a lot for real coding workflows:

Reactive Delta Compaction

View all

ContextGraph

Memory backend for coding agents

Memory backend for coding agents

Forum Threads

You do not have to lose too much working state when context gets tight.

Forum Threads

You do not have to lose too much working state when context gets tight.