master

huyubiaofix(memory-cases): do not automatically rebuild databases

7f4bf47f创建于 14 天前历史提交

文件	最后提交记录	最后更新时间
http-mcp-deployment.md	fix(memory-cases): do not automatically rebuild databases Keep missing business and index databases visible instead of silently recreating empty files during runtime operations. - create the business DB only at startup and require existing runtime DBs - return distinct missing database errors through REST and MCP - stop polling from recreating databases and rate-limit failure logs - rebuild the index asynchronously with validation and atomic swap - reject SQLite URI paths and document manual recovery behavior Co-Authored-By: GPT-5.6 Sol	14 天前
locomo-evaluation.md	chore: restructure into ram-a-mem subdir and scaffold ram-a-kv Move all project files (Cargo workspace, crates, docs, evaluation, plugins, LICENSE, README, Third-Party notice) into ram-a-mem/ via git mv to prepare the repo for hosting a second project. Keep .gitignore and .gitattributes at the repo root and re-anchor .gitignore paths to the ram-a-mem/ prefix so ignore rules remain effective. Add an empty ram-a-kv/ as a scaffold for the upcoming project.	1 个月前
memory-cases-manual-quick-verify.md	chore: restructure into ram-a-mem subdir and scaffold ram-a-kv Move all project files (Cargo workspace, crates, docs, evaluation, plugins, LICENSE, README, Third-Party notice) into ram-a-mem/ via git mv to prepare the repo for hosting a second project. Keep .gitignore and .gitattributes at the repo root and re-anchor .gitignore paths to the ram-a-mem/ prefix so ignore rules remain effective. Add an empty ram-a-kv/ as a scaffold for the upcoming project.	1 个月前
memory-cases-qa-evaluation.md	chore: restructure into ram-a-mem subdir and scaffold ram-a-kv Move all project files (Cargo workspace, crates, docs, evaluation, plugins, LICENSE, README, Third-Party notice) into ram-a-mem/ via git mv to prepare the repo for hosting a second project. Keep .gitignore and .gitattributes at the repo root and re-anchor .gitignore paths to the ram-a-mem/ prefix so ignore rules remain effective. Add an empty ram-a-kv/ as a scaffold for the upcoming project.	1 个月前
model-compatibility.zh-CN.md	merge: resolve PR 24 conflicts and address review	15 天前
ram-a-mem-configuration-and-pipeline.zh-CN.md	fix(memory-cases): do not automatically rebuild databases Keep missing business and index databases visible instead of silently recreating empty files during runtime operations. - create the business DB only at startup and require existing runtime DBs - return distinct missing database errors through REST and MCP - stop polling from recreating databases and rate-limit failure logs - rebuild the index asynchronously with validation and atomic swap - reject SQLite URI paths and document manual recovery behavior Co-Authored-By: GPT-5.6 Sol	14 天前
ram-a-mem-configuration-reference.zh-CN.md	fix(ram-a-mem): address security audit findings (issue #9) Incorporate the follow-up review fixes for duplicate grounding IDs, fixed-endpoint redirect handling, strict localhost validation, and case-import worker failure propagation. Co-Authored-By: GPT-5.6 Sol	15 天前
ram-a-mem-prerequisites.zh-CN.md	docs(ram-a-mem): add user-facing prerequisites guide Collect the external conditions a ram-a-mem deployment must prepare into a single guide: the minimal viable setup (one OpenAI-compatible Chat endpoint whose final answers land in choices[0].message.content), the hash-vs-real embedding boundary, per-feature model requirements (rerank, graph memory, case library), MCP client requirements, and network and deployment boundaries. Installation artifacts (binary, delivered JSON config, RPM layout) are intentionally out of scope. Co-Authored-By: Claude Code <noreply@anthropic.com>	14 天前
ram-a-mem-rpm-agent-self-test.zh-CN.md	fix(ram-a-mem): address follow-up review on docs and strict schema - Update the configuration guide's error mapping table to the actual codes: EMBEDDING_FAILED / SQLITE_BUSY / SQLITE_READONLY have their own entries and retriable values; STORAGE_FAILED is documented as the fallback only. - Complete the nested object definitions (subject, object, event_time, attributes) in the extraction schema so strict json_schema validation accepts the whole schema, pinning only the fields the pipeline reads. - Make the RPM self-test guide verify downloaded RPMs against an expected sha256 (RAM_A_RPM_SHA256 / XIAOO_RPM_SHA256 with sha256sum --check) instead of only recording the digest. Co-Authored-By: Claude Code <noreply@anthropic.com>	22 天前
xiaoo-case-library-integration.md	fix(memory-cases): do not automatically rebuild databases Keep missing business and index databases visible instead of silently recreating empty files during runtime operations. - create the business DB only at startup and require existing runtime DBs - return distinct missing database errors through REST and MCP - stop polling from recreating databases and rate-limit failure logs - rebuild the index asynchronously with validation and atomic swap - reject SQLite URI paths and document manual recovery behavior Co-Authored-By: GPT-5.6 Sol	14 天前