{"slug": "what-you-save-when-project-context-stops-repeating", "title": "What you save when project context stops repeating", "summary": "Qarinah, a local-first project-memory tool, compiles a compact, cited memory pack instead of replaying full project history to coding agents. In benchmarks across six software-task fixtures, it reduced estimated input-context tokens from 442,113 to 5,682 while maintaining direct coverage of all required targets in the top five results, and passed 380 of 380 deterministic file-specific queries. The project is Apache-2.0 and integrates with Codex, Claude Code, Cursor, CLI tools, and MCP clients.", "body_md": "Qarinah compiles a compact, cited project-memory pack instead of asking every new coding-agent session to replay the entire available history.\n\nAcross six committed software-task fixtures, the full-history baseline contained **442,113 portable estimated input-context tokens**. The Qarinah path used **5,682**. Every required target was still directly covered in the top five results.\n\nThat is:\n\nThe ratio is not a claim that every provider bill drops by 98.71%, or that an agent session lasts 77.81 times longer. It measures the compared input-context volume in the published six-fixture estimate.\n\nThe table applies four flat, uncached input-token rates to the same two token estimates. It is arithmetic, not a provider invoice.\n\n| Flat uncached input rate | Full-history baseline | Qarinah pack | Estimated saving |\n|---|---|---|---|\n| $1 / million tokens | $0.442113 | $0.005682 | $0.436431 |\n| $3 / million tokens | $1.326339 | $0.017046 | $1.309293 |\n| $5 / million tokens | $2.210565 | $0.028410 | $2.182155 |\n| $15 / million tokens | $6.631695 | $0.085230 | $6.546465 |\n\nThe calculation is:\n\n`estimated tokens / 1,000,000 x flat input rate`\n\nIt deliberately excludes provider-native tokenization, caching, output tokens, reasoning tokens, tool calls, retrieval, hosting, and fixed fees. Real cost depends on the provider, model, cache behavior, context composition, and how often the same history would otherwise be resent.\n\nCompression only matters if the next task can still find its evidence. The benchmark checks both volume and retrieval coverage: every required target had to be directly present in the top five. Qarinah preserves the source event ID and content hash for selected context, so a later agent receives a bounded handoff that can be inspected instead of an opaque story.\n\nQarinah also passed **380 of 380 deterministic file-specific exact and typo-tolerant queries** across 40-, 50-, and 100-file projects. Those tests verify retrieval behavior; they do not establish universal task quality or a ranking against every memory system.\n\nThe methodology, fixture records, costs, and exclusions are public:\n\nQarinah is Apache-2.0 and local-first. It works with Codex, Claude Code, Cursor, CLI tools, and compatible MCP clients. Project capture is opt-in, metadata-only by default, and does not capture hidden reasoning.\n\n**Would you trust agent memory more if every selected item pointed back to its source?**", "url": "https://wpnews.pro/news/what-you-save-when-project-context-stops-repeating", "canonical_source": "https://dev.to/ajnasnb7717/what-you-save-when-project-context-stops-repeating-19ao", "published_at": "2026-08-11 03:20:45+00:00", "updated_at": "2026-08-11 03:45:37.629688+00:00", "lang": "en", "topics": ["developer-tools", "ai-tools", "ai-agents", "large-language-models"], "entities": ["Qarinah", "Codex", "Claude Code", "Cursor", "MCP"], "alternates": {"html": "https://wpnews.pro/news/what-you-save-when-project-context-stops-repeating", "markdown": "https://wpnews.pro/news/what-you-save-when-project-context-stops-repeating.md", "text": "https://wpnews.pro/news/what-you-save-when-project-context-stops-repeating.txt", "jsonld": "https://wpnews.pro/news/what-you-save-when-project-context-stops-repeating.jsonld"}}