Agent memory

Persistent context for agents that learn over time.

Red ocean · Measured classification

Agent memory: Cooling recently, still above last year

Search attention for “ai agent memory” is falling. 12 reviewed open-source projects serve this task. Recent search attention is cooling while staying above last year's level. Use interviews and product usage to assess customer demand.

Report dated 2026-10-15 · Method 2.0.0 · moderate evidence confidence

Measured search term: ai agent memory · Last 8 complete weeks vs previous 8

2609 active projects match the published GitHub search scope.
The keyword “ai agent memory” has 91 complete weeks; 1% are reported as zero.

Open-source competition: ≥67 / 100 · Established alternatives
Open-source alternatives: 12
Direction basis: Recent search windows
Matching active GitHub projects: 2,609
Search-interest growth: -46% · Last 8 complete weeks vs previous 8
Search term and region: ai agent memory · Worldwide
Complete weekly observations: 91

Why this classification

Recent search attention is cooling while staying above last year's level. Both horizons provide context.
Median weekly search interest fell 46% across two consecutive eight-week windows.
2,609 active repositories match this search scope. Review their users and workflows to identify direct alternatives.
Four-week search change: 7%; thirteen-week change: -71%. These windows check the direction of the eight-week comparison.
Project roles: 12 direct alternatives, 0 adjacent projects, 5 resources, 83 awaiting review.
Competition pressure: 67–100/100; breadth 26, established alternatives 25, leading project strength 16.

Source evidence

GitHub repository search · topic:agent-memory fork:false archived:false stars:>=1 pushed:>=2025-10-15

Collected 2026-10-15T05:45:57.667Z

Google Trends search interest · Collected 2026-10-15T05:45:57.667Z

Leading repositories

Repository	Stars	Description
666ghj/MiroFish	79,281	A Simple and Universal Swarm Intelligence Engine, Predicting Anything. 简洁通用的群体智能引擎，预测万物
bojieli/ai-agent-book	54,288	《深入理解 AI Agent：设计原理与工程实践》（李博杰著）开源主仓库：全书正文、编译版 PDF 与按章配套代码
pingcap/tidb	40,642	TiDB is built for agentic workloads that grow unpredictably, with ACID guarantees and native support for transactions, analytics, and vector search. No data silos. No noisy neighbors. No infrastructure ceiling.
volcengine/OpenViking	39,949	Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
topoteretes/cognee	32,312	Cognee is the open-source AI memory platform for agents. Give your AI agents persistent long-term memory with small models for free
supermemoryai/supermemory	31,229	Memory and context engine + app that is extremely fast, scalable, and can be run fully locally. The Memory API for the AI era.
dolthub/dolt	24,640	Dolt – Git for Data
MemoriLabs/Memori	17,147	Memori is agent-native memory infrastructure. A LLM-agnostic layer that turns agent execution and conversation into structured, persistent state for production systems. Built for enterprise, Memori works with the data infrastructure you already run, no rip-and-replace, and deploys across managed cloud, single-tenant cloud, VPC, and on-premises.
NevaMind-AI/memU	14,546	Personal memory across agents
semantica-agi/semantica	14,044	Graph-Native Infrastructure for Context and Accountable AI Systems

Limits of this result

Google Trends measures relative search attention. Assess customers, revenue and willingness to pay through direct research.
Supply covers active repositories matching the displayed GitHub queries. Broader competition research can include other repositories and commercial products.
Repository density and search attention provide complementary research signals. Assess competition and demand through user and product evidence.
Competition pressure is an operational index of observed open-source alternatives. Stars and forks indicate developer attention and reuse; user adoption and commercial products deserve separate research.
The displayed projects form a stars-ranked sample. Competition pressure is a lower bound; wider coverage can strengthen the assessment.
Some project roles await closer review. The pressure range includes their possible contribution.

Search interest measures attention. Customer demand needs user research; classification thresholds are published heuristics subject to empirical calibration.

Community requests

[Feature]: Add Kilo Code memory integration (ported from opencode-plugin)

Discussion open

Issue Origin Observed or reproduced in a real environment Problem Statement OpenViking's harness support matrix currently covers claude-code, codex/trae-cli, cursor, trae/trae-cn, zcode, opencode, dsh, pi, openclaw, and hermes (see docs/en/agent-integrati

volcengine/OpenViking · Posted: 2026-09-19 · Updated: 2026-09-21 · Collected: 2026-10-15

[Feature]: dsh-memory-plugin — session filter to skip capture/recall for subagent & derived sessions

Discussion open

Problem This plugin hooks every session unconditionally (inject = ["agents", "sessions", "tools"]): - agent/session-start → injectStartupProfile injects the OpenViking profile into subagent sessions too (subagent / subagentfork / workflow children, AgentTea

volcengine/OpenViking · Posted: 2026-08-20 · Updated: 2026-10-15 · Collected: 2026-10-15

[Feature]: Refactor VectorDB storage abstraction to support self-hosted external backends such as PostgreSQL + pgvector

Discussion open

Problem Statement - OpenViking's current openviking/storage/vectordb and openviking/storage/vectordbadapters layers are still tightly coupled to the existing collection runtime and index lifecycle. - This makes it difficult to integrate with self-hosted exte

volcengine/OpenViking · Posted: 2026-06-01 · Updated: 2026-07-18 · Collected: 2026-10-15

feat(hermes-plugin): allow capture and explicit tools without automatic recall

Discussion open

Issue Origin Observed or reproduced in a real environment The missing control was identified by inspecting the Hermes plugin and reproduced locally through Hermes's external-provider discovery in temporary profile homes. With autorecall: false in YAML or OP

volcengine/OpenViking · Posted: 2026-10-14 · Updated: 2026-10-15 · Collected: 2026-10-15

[Feature]: Add official Qwen Code host adapter to agent-hook-plugin (hosts/qwen)

Discussion open

Issue Origin Observed or reproduced in a real environment Problem Statement agent-hook-plugin currently ships official adapters and one-line installers for several hosts (Cursor, Kimi Code, Trae, ZCode, DSH, ...), but not for Qwen Code (https://github.com

volcengine/OpenViking · Posted: 2026-10-15 · Updated: 2026-10-15 · Collected: 2026-10-15

[Feature]: zcode plugin — capture subagent work products into memory (opt-in)

Discussion open

Issue Origin: Observed or reproduced in a real environment Feature Area: Session Management Problem Statement Work performed by ZCode subagents never reaches OpenViking memory. The ZCode adapter captures a session at Stop by parsing the rollout file (hosts

volcengine/OpenViking · Posted: 2026-10-14 · Updated: 2026-10-15 · Collected: 2026-10-15

deja recall is missing from zsh completion, and Tab offers it search flags it refuses

Discussion open

deja recall takes --project, --harness and --limit (cmd/deja/recallcli.go:16). No shell has a case for it, so it falls through to the search flags, and zsh does not offer the command at all: its command list is hand-written and recall was never added. $ deja

vshulcz/deja-vu · Posted: 2026-10-14 · Updated: 2026-10-14 · Collected: 2026-10-15

Use and share the evidence

Permanent report · Markdown · JSON · PNG card

Use the open-source CLI and MCP server on GitHub