{"slug": "i-accidentally-hit-sota-on-agentic-memory-by-using-ai-companions", "title": "I accidentally hit SOTA on agentic memory by using AI companions", "summary": "Developer tools startup graphCTX claims its local memory system for AI coding agents achieves state-of-the-art performance, retrieving repo facts in ~1ms with flat latency even at 5,000 facts, outperforming cloud-based alternatives like Supermemory. The tool captures coding facts from repos, binds them to git state, and promotes only durable knowledge, aiming to reduce context repetition for developers.", "body_md": "#\nBuilt to be the best\n\ncontext layer for AI coding agents.\n\ngraphCTX keeps repo knowledge close to the work: commands, conventions, decisions, and hard-won fixes. Developers spend less time re-explaining context and more time shipping, with local memory that is fast, private, and reliable.\n\n`curl -fsSL https://graph.coder.company/install | sh`\n\n[See the benchmarks →](#bench)\n\nuser: what's the deploy command?\n\nagent: I don't see one in the repo —\nyou'll need to tell me.\n\nuser: what's the deploy command?\n\nagent: ./scripts/ship.sh --canary --wait\nmem:9f3a2c\n\nSame agent, same prompt. graphCTX gives it the repo memory developers otherwise repeat.\n\n## How graphCTX manages memory.\n\nThe system is deliberately narrow: capture reliable coding facts, keep them valid as the repo changes, and return only the context the agent can use.\n\n### Extract trusted coding facts\n\ngraphCTX reads package scripts, lockfiles, CI, editorconfig, AGENTS.md, and session episodes so the memory base starts from repo evidence, not model guesses.\n\n### Bind memory to git state\n\nFacts are valid against commits and branches. When code changes, memory moves with the DAG instead of drifting on wall-clock timestamps.\n\n### Promote only durable knowledge\n\nSession details can become workspace or user memory only after evidence gates. Secrets and low-trust prose stay out of durable context.\n\n### Choose the smallest useful set\n\nA relevance gate scores topic drift, entities, and file scope so the agent gets specific context instead of a noisy memory dump.\n\n### Attach provenance to every recall\n\nReturned memory is compact, budgeted, and tagged with source provenance, making it easier for developers to trust and audit what the agent uses.\n\n## Benchmarked against Supermemory.\n\nSame coding-fact set, same queries. graphCTX runs locally and answers in\n~1ms; a live Supermemory search round-trip measured ~494ms (p50), so\nrecall stays fast and predictable during development.\nReproducible: `graphctx compare --deep`\n\n.\n\n### Latency stays flat at scale\n\nPer-prompt retrieval p50/p95 as the workspace grows. Indexed lookup plus a bounded semantic re-rank keeps the hot path at ~1ms — a 5,000-fact monorepo retrieves as fast as an empty one.\n\n### Reliable after compaction\n\nPost-compaction solve rate across 14 coding tasks. graphCTX restored the needed repo fact in every run.\n\n// graphctx eval run --arms A,B,C\n\n// scope: this compares local latency + cost on direct coding-fact retrieval for developer workflows. Supermemory targets general/conversational memory with cloud connectors, cross-document reasoning, and neural embeddings that graphCTX doesn't attempt.\n\n## Start using repo memory in 30 seconds\n\nNo account. No API key. No cloud setup. Install the CLI, connect your agent, and give every session repo-aware memory.\n\n`curl -fsSL https://graph.coder.company/install | sh`\n\nPrefer npm? npm i -g graphctx\n\n` $ curl -fsSL https://graph.coder.company/install | sh `\n\ninstall the CLI (detects Node / Bun)\n\n` $ graphctx install claude `\n\nwire your agent (claude · cursor · opencode · generic)\n\n` $ graphctx doctor `\n\nverify graphCTX is connected", "url": "https://wpnews.pro/news/i-accidentally-hit-sota-on-agentic-memory-by-using-ai-companions", "canonical_source": "https://graph.coder.company/", "published_at": "2026-06-14 07:03:45+00:00", "updated_at": "2026-06-14 07:29:45.636245+00:00", "lang": "en", "topics": ["ai-tools", "developer-tools", "ai-agents", "large-language-models"], "entities": ["graphCTX", "Supermemory", "Claude", "Cursor", "OpenCode"], "alternates": {"html": "https://wpnews.pro/news/i-accidentally-hit-sota-on-agentic-memory-by-using-ai-companions", "markdown": "https://wpnews.pro/news/i-accidentally-hit-sota-on-agentic-memory-by-using-ai-companions.md", "text": "https://wpnews.pro/news/i-accidentally-hit-sota-on-agentic-memory-by-using-ai-companions.txt", "jsonld": "https://wpnews.pro/news/i-accidentally-hit-sota-on-agentic-memory-by-using-ai-companions.jsonld"}}