Tencent Team Memory: Agents Bir Memory Paylaşıyorsa Birinin Owner Olması Gerekir

Tencent'ın latest open-source release için easy read data hakkında. Chinese cloud vendor team'in collective memory'sini tutan system ship ediyor, headline'ın kalanını kendiniz yazabilirsiniz. Aynı zamanda wrong read, actual repository ile beş dakika geçirmek bunu aşmaya yeter.

13 Ağustos'ta Tencent Cloud, agent memory'yi personal convenience'dan shared team infrastructure'a taşıyan TencentDB Agent Memory major update Team Memory'yi announced. Thing'i run etmek yerine announcement ve repo üzerinden gittim, yani desk read, field report değil. What it is, "team" neden real story ve Western readers'ın first reach worry neden wrong place burada.

Temel çıkarımlar - Team Memory TencentDB Agent Memory'yi individual long-term memory'den shared team assets'e genişletiyor: chat history, queryable wiki olarak documents, graph olarak code repositories, distilled reusable skills, hepsi access controls behind. - İsme rağmen database değil. SQLite üzerinde vector search kullanan local-first TypeScript services suite, MIT-licensed, Anthropic own API protocol konuşarak Claude Code'a plug. - Default deployment everything own environment'ta tutar, reflexive "my data goes to China" worry mostly apply etmez. Residual risks governance, Chinese-first community, optional Tencent Cloud VectorDB path. - Shared memory shared attack surface. Memory-poisoning research corrupted entries sessions across propagate diyor, burada teammates' agents across by design propagate eder. - Vendor benchmark claims self-reported. GitHub traction, 27 Ağustos'ta 24.900+ stars, real, fakat Western developer uptake number'dan much thinner görünüyor.

Ne oldu

13 Ağustos 2026 Tencent Cloud, May'de open-sourced TencentDB Agent Memory'nin second major release'i Team Memory'yi announced. Announcement ve repository'den key facts:

• Team memory four asset types'tan assembled: Chat Memory, historical agent conversations; LLM-Wiki, natural language queryable project documents; Code Graph, repository structure, symbols, call paths; Skills, reuse için distilled completed troubleshooting session veya code review.

• New console Memory Hub teams, agents, tasks manage eder, every memory item owner, version, status, usage history taşır. Permissions per user, role, agent, private'dan team-wide'a.

• Assets role by role assembled: bug-fixing agent Code Graph ve past troubleshooting skills; requirements agent wiki ve business context.

• Tencent CodeBuddy, OpenClaw, Claude Code dahil agent platforms across çalışır; code release v2.0.0 3 Ağustos'ta landed, newsroom post'tan ten days before.

Keep claim portability: "the accumulated experience in Team Memory remains fully compatible even if the underlying models or agent frameworks change." Tencent project first 90 days 20.000 GitHub stars geçti, GitHub Trending more than once top diyor. Star count today verifiable, 27 Ağustos kontrolümde 24.912, third-party tracker 4 July'da 7.200 recorded, trajectory plausible. Trending claim Tencent post'a traces ve check yok, marketing file.

Tencent Cloud Team Memory release open-source TencentDB Agent Memory project'i individual'dan team use'a genişletiyor, conversations, documents, code, distilled skills'i Claude Code dahil platforms için governed memory assets olarak organize ediyor, 13 Ağustos 2026 announcement'a göre. Repository 27 Ağustos 2026'da 24.912 stars, GitHub'a göre.

Practitioner read: actually nedir

First, name. TencentDB Agent Memory database değil, PostgreSQL built değil, meaningful sense TencentDB technology değil. MemoryCore, MemoryKnowledge, MemoryPanel, MemoryProxy TypeScript services set'i, local SQLite + sqlite-vec vector search, scale-out için optional Tencent Cloud VectorDB path. TencentDB label brand adjacency. Tencent database team genuinely infrastructure open-sourced before, TBase, now OpenTenBase, 2019, fakat bu org'dan application-layer project.

Architecture star-chasing'in düşündüğünden considered. Memories raw conversation'dan stable persona-level facts'a layers through distilled; retrieval BM25 keyword search + vector search, item/time budgets altında, injected context small. Cleverest piece Memory Proxy: agent ile model arasında, Anthropic/OpenAI API protocols ikisini de konuşur, relevant memory system prompt'a inject. Claude Code için no plugin, no hook, no MCP server. Client never knows.

Real software da. One-command Docker deployment, TypeScript/Python SDKs, OpenAPI docs, bilingual documentation, daily commits. Default branch feat/server_team, main değil, still fast moving.

TencentDB Agent Memory local-first TypeScript service suite, SQLite + sqlite-vec storing, layered memory distillation, hybrid BM25-plus-vector retrieval. Memory Proxy Anthropic/OpenAI API protocols konuşur, Claude Code plugin/MCP server olmadan long-term memory alır, project README'ye göre, retrieved 28 Ağustos 2026.

"Team" neden real story

Agent memory category mostly single-player. Mem0 best-funded entrant, October 2025 $24 million raised, AWS agent SDK memory provider. Zep Graphiti temporal knowledge graph, facts invalidated not deleted. Letta every agent'a self-managed tiered memory. All three one agent remembering own past.

En çok watched dört open-source agent memory project GitHub stars, 27-28 Ağustos 2026 checked: Mem0, Graphiti, TencentDB Agent Memory, Letta. Tencent entry years by youngest.

Team Memory bet unit of memory team olacak, agent değil. Real work agents kullanan anyone problem bilir: project background every session re-explained, last month fix nobody reconstruct, working method one person's Claude Code history. Tencent answer all managed assets with owners, versions, permissions; different agent roles different slices.

Last part genuinely new. "Private" team admins bile item read edemez, developer distilled troubleshooting skill reviewed sonra specific agents/people share edilebilir access-control model, memory as team infrastructure. Western projects core olarak ship etmiyor. Better notebook satıyorlar; bu shared filing cabinet with lock policy istiyor.

Mem0, Zep, Letta single agent remembering own history centre. Team Memory conversations, documents, code graphs, distilled skills'i per-user/per-role/per-agent permissions olan governed team assets, task assembled olarak treat ediyor, Tencent Cloud announcement ve repository documentation'a göre.

Shared brain shared target

Announcement'ın dwell etmediği. Memory bigger context windows retention solve etmediği için var: Chroma context rot study 18 frontier models input grows oldukça degrade, window fill before. Everyone memory build, security research delicate reason caught up. OWASP agentic threat guidance memory poisoning outright: corrupt long-term store once, every future session inherits corruption.

Store shared yapın. Poisoned troubleshooting skill next week own agent'ı değil, shared every agent/teammate'i by design mislead. Team conversations, documents, code structure, working methods repository outsider için tidy attacker catalogue. Tencent mitigations real but procedural: ACL who can read what limit, skills sharing before reviewed, human in loop machine-distilled content. Control review habit kadar strong.

Use etmeme reason değil. Governance features product olmasının reason, "which agent may read which memory" stack'teki any access policy kadar serious, agent governance generally gibi.

Agent security research, OWASP agentic AI threat guidance dahil, memory poisoning'i corrupted long-term memory future sessions steers distinct attack class olarak identifies. Shared team memory store risk'i teammates/agents across by design extends, review-before-share workflows ve access control load-bearing features.

China question, honestly answered

Deployment assumption first, answer tamamen değişir. Open-source release'i self-host, MIT-licensed/local-first, data own environment'ta. Storage local SQLite, LLM endpoints yours configure, no Tencent Cloud account. Bu path'te Chinese vendor holding team memory reflex worry apply etmez, default deployment için wrong worry.

Real residual concerns quieter. Optional Tencent Cloud VectorDB path use edersen Tencent cloud tie, separate decision. Repo OpenTelemetry instrumentation includes, deploy before normal hygiene audit, accusation değil. Community Chinese-first: discussions, tutorials, momentum. Enterprise support istersen Chinese vendor'dan, bazı organisations procurement conversations.

One number community shape star count'tan better: just under 25k GitHub stars; earlier this month Hacker News submission 2 points no comments. Stars real, centre domestic. Western teams için help where caveat. Chinese teams going global için tool burada teams work way built hint, own build-versus-buy consideration.

MIT-licensed TencentDB Agent Memory self-hosting data'yı own environment local SQLite'ta tutar, user-configured model endpoints, no Tencent Cloud account required, repository documentation'a göre. Residual considerations optional Tencent Cloud VectorDB path, standard dependency auditing, Chinese-first community/support channel.

Şimdi ne yapmalı

Agent memory radar'daysa three steps order.

1. Bugün: Announcement değil repo read. Architecture docs/ACL actual design, Docker one command.

2. Bu hafta: One non-critical repository + one agent against stand up, distillation layers actually retain watch. Vendor benchmarks, Tencent PersonaMem 48% to 76% reports, independently unreproduced, own corpus substitute değil.

3. Bu ay: Anything real import before access policy write. Which memories private/team-wide, distilled skill before shared who reviews. Tool controls gives; policy yours.

Two things don't. Policy before proprietary code/client conversations shared store import etmeyin, retroactive permissioning worse than no memory. Star count Western readiness signal değil; debugging community mostly Chinese. Earlier journey agents fit weighing iseniz agentic architecture piece start.

FAQ

TencentDB Agent Memory database mi?

No. Name rağmen TypeScript services suite, local SQLite + vector search, agents ve models arasında. TencentDB label build team, technology değil. Optional Tencent Cloud VectorDB integration scale-out, default no database server.

TencentDB Agent Memory data Tencent Cloud'a gönderir mi?

Default self-hosted setup'ta no. MIT-licensed, local SQLite, configure ettiğiniz LLM endpoint. No Tencent Cloud account. Exception optional Tencent Cloud VectorDB, deliberate choice not default.

Claude Code ile çalışır mı?

Yes, neat. Memory Proxy Anthropic/OpenAI API protocols konuşur, Claude Code, CodeBuddy, OpenClaw system prompt üzerinden injected memory, no plugin/hook/MCP server. Agent proxy'ye point.

Mem0 ile nasıl compare?

Mem0 more established Western option: older, better funded, October 2025 $24M raised, AWS agent SDK integrated. TencentDB younger, local-first default, team-level governance distinguishing: owners, versions, permissions, role-based assembly. One agent remembering problem either works. Team sharing memory safely problem Tencent built around.

Sonuç

Team Memory young fast-moving, databases ile better known team, vendor benchmarks/Trending claims usual discount. Design instinct right: agents real team work yaptığında memory personal convenience değil shared infrastructure, owners, permissions, review needs. Tencent open-source projects arasında first, own hardware run no send anyone. Watch question star count değil. Review-before-share discipline hurried teams ile survive mı, works veya quietly liability burada.

Shared agent memory own stack fit çalışıyorsanız clients ile regular conversation. İletişime geçin.

Kaynaklar

• Tencent Cloud, "TencentDB Agent Memory Releases Team Memory": https://www.tencentcloud.com/dynamic/news-details/101465 (yayımlanma 2026-08-13, erişim 2026-08-28)

• TencentCloud, TencentDB-Agent-Memory repository: https://github.com/TencentCloud/TencentDB-Agent-Memory (erişim 2026-08-28; star and fork counts checked 2026-08-27)

• MarkTechPost, "Tencent Cloud Open Sources TencentDB Agent Memory v2.0": https://www.marktechpost.com/2026/08/07/tencent-cloud-open-sources-tencentdb-agent-memory-v2-0/ (yayımlanma 2026-08-07, erişim 2026-08-28)

• Open Source For You, "Tencent Cloud Agent Memory v2": https://www.opensourceforu.com/2026/08/tencent-cloud-agent-memory-v2/ (yayımlanma 2026-08, erişim 2026-08-28)

• PR Newswire via Morningstar, "Mem0 raises $24M": https://www.morningstar.com/news/pr-newswire/20251028sf07039/mem0-raises-24m-series-a-to-build-memory-layer-for-ai-agents (yayımlanma 2025-10-28, erişim 2026-08-28)

• Mem0 repository: https://github.com/mem0ai/mem0 (erişim 2026-08-28)

• Zep, Graphiti repository: https://github.com/getzep/graphiti (erişim 2026-08-28)

• Letta repository: https://github.com/letta-ai/letta (erişim 2026-08-28)

• Chroma, "Context Rot: How Increasing Input Tokens Impacts LLM Performance": https://research.trychroma.com/context-rot (yayımlanma 2025-07, erişim 2026-08-28)

• OWASP, "Agentic AI Threats and Mitigations": https://genai.owasp.org/resource/agentic-ai-threats-and-mitigations/ (erişim 2026-08-28)

• OpenTenBase (formerly TBase): https://www.opentenbase.org/en/ (erişim 2026-08-28)

• Wikimedia Commons, cover image "African Bush Elephant" (GFDL 1.2, Muhammad Mahdi Karim): https://commons.wikimedia.org/wiki/File:African_Bush_Elephant.jpg (erişim 2026-08-28)