▲РЕПО НЕТ В ОКНЕ 30D. ПОКАЗАН FINDING ИЗ ОКНА 180D (180 days) — РАНГ #14.
FINDING #14 · UNIT ID 1164477708
mksglu/context-mode
Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks.
SURPRISE SCORE
0.00
Score Breakdown
SCORE = SURPRISE × ENGAGEMENT^0.7 × FRESHNESS × VISIBILITY × CONFIDENCE
SURPRISE = WINDOW STARS / DAYS / (AUDIENCE + FLOOR)
14% OF STARS IN ARCHIVE
Growth Telemetry
VELOCITY /D
14.88
ACCEL
-0.18
RETENTION
5.4%
PEAK 2026-03-01 · FORK-RETENTION 93.3% · 2,678 STARS / WINDOW
Author Audience
AUDIENCE
2,534
FOLLOWERS
530
OWNER ★
20,044
Engagement Signals
FORKS
1,401
ISSUE AUTH
85
PR AUTH
73
UNIQUE STARGAZERS 2,675 / 2,678 (DIVERSITY 1.00)
Потенциал заработка
C48/100
СПРОС88
Деньги в теме есть: AI code assistants market оценивается в $8B–$10B+ в 2025/2026 и растёт 20–48% CAGR по разным источникам.
ЗАХВАТ50
Value clear, но Cursor, Copilot, Claude Code, Codex, JetBrains, Tabnine и Augment могут встроить context/memory/analytics нативно; OSS-аналогов много.
ДОСТУП38
GitHub enrichment даёт SPDX None, package/npm declares Elastic-2.0; hosted/managed competing service restricted, поэтому сторонний SaaS-клон юридически слаб.
«Потенциал топят license/access и слабый defensibility против владельцев AI IDE/agent платформ.»
Рыночный анализ · Обзор
Context Mode — MCP-сервер и hooks для AI coding agents, которые экономят context window и сохраняют память сессии.
AI developer tooling / MCP / agent memory / context optimization.ЗРЕЛОСТЬ · РАСТЁТ
ЯЗЫК
TypeScript
ЛИЦЕНЗИЯ
НЕТ · ALL RIGHTS RESERVED
РЕЕСТР
npm
РЕЛИЗ
v1.0.169
КОНТРИБЬЮТОРЫ
100
УСТАНОВКА: npm install context-mode
Что делает
Context Mode перехватывает тяжёлые вызовы инструментов, выполняет код в sandbox и не сбрасывает сырые логи, snapshots и файлы прямо в чат. Он сохраняет события сессии в SQLite + FTS5 и даёт агенту искать релевантное через BM25 после compaction. Проект поддерживает Claude Code, Gemini CLI, Copilot, Cursor, Codex CLI, OpenCode, Zed и другие клиенты через MCP, hooks и routing rules.
Какую боль решает
AI coding agents быстро забивают context window, забывают ход работы после compaction и тратят деньги на повторное чтение тех же файлов, логов и issue.
Сценарии использования
- Длинная Claude Code/Codex/Cursor-сессия над большим repository без потери контекста после compaction.
- Анализ логов, GitHub issues и Playwright snapshots без вставки десятков KB в prompt.
- Командная аналитика AI-assisted engineering через ctx_insight.
- Единые routing rules для 10+ agent clients в компании.
- Локальный persistent memory слой для задач, ошибок, файлов и git-событий.
Целевой пользователь
Power users AI coding agents, platform/devtools teams, engineering managers и команды с дорогим Claude/Cursor/Codex usage.
Open-source аналоги
Даёт актуальную документацию в prompt; не решает session memory и tool-output sandbox так глубоко.
Универсальная memory layer для agents/apps, Apache-2.0 и managed service; не заточен только под coding hooks.
Semantic code retrieval/editing tools через MCP; больше про IDE-level code intelligence, меньше про context-budget accounting.
Stateful agents с advanced memory; шире и платформеннее, но не такой лёгкий drop-in для 17 coding clients.
Temporal knowledge graph для agent memory; сильнее graph/RAG, слабее в client hooks/routing.
Очень близко: persistent memory for AI coding agents, hooks/MCP/REST.
Local-first event-sourced memory + judgment layer; MIT, plain JSONL.
Позиционирование
Context Mode — не лидер всего OSS-контекста, но сильный нишевый догоняющий/лидер в узкой подкатегории “context budget + sandboxed tool output + cross-client routing for coding agents”. Комбинация hooks + SQLite/FTS5 + routing enforcement + savings accounting выглядит дифференцированной, не клоном.
Коммерческие аналоги
AI coding assistant, agent mode, code review и enterprise controls.
Microsoft reported 26M+ Copilot users, 4.7M paid Copilot subscribers и 140k organizations using GitHub Copilot.
Individual Pro$10/mo
Individual Pro+$39/mo
Individual Max$100/mo
Business$19/user/mo
Enterprise$39/user/mo
AI-native IDE with agents, MCP, skills, hooks, cloud agents and team analytics.
Reported >$2B ARR; Bloomberg reported raise talks around >$50B valuation.
Pro$20/mo
Pro+$60/mo
Ultra$200/mo
Teams$40/user/mo
Enterprisecustom
Terminal/IDE coding agent that edits files, runs commands and handles long tasks.
Claude Code reached $1B run-rate in 6 months; later Anthropic stated Claude Code run-rate revenue >$2.5B.
Claude Team standard monthly$25/user/mo
Claude Team standard annual$20/user/mo
Claude Codeincluded in Claude paid plans
API Opus 4.7 input$5/MTok
API Opus 4.7 output$25/MTok
AI IDE with Cascade agent, codebase context and deploy previews.
Exact revenue/users not confirmed.
Pro$15/mo
Teams$30/user/mo
Extra Pro credits$10/250 credits
Extra Team credits$40/1000 credits
Enterprisecustom
Private AI code assistant, agentic platform and context engine deployable as SaaS/VPC/on-prem/air-gapped.
Tabnine claims millions of developers and thousands of companies.
Code Assistant$39/user/mo annual
Agentic Platform$59/user/mo annual
LLM provider usage handling feeprovider usage + 5%
Context Engine standalone$5,800 enterprise price shown
Coding agent + context engine + MCP/native tools + remote compute.
Exact revenue/users not confirmed; pricing page positions product for teams/enterprise.
Business$100/mo flat, up to 50 seats, $100 usage included
LLM usageprovider list price + 40% service fee
Enterprisecustom
AWS developer assistant, coding agent, transformation, security and troubleshooting.
AWS distribution, exact Q Developer users not confirmed.
Free$0, 50 chat interactions/mo
Pro$19/user/mo, 1,000 chat interactions/mo
Transformation agents4,000 LOC/mo/user included
IDE-native AI assistant and coding agent inside JetBrains IDEs.
Exact AI users not confirmed; JetBrains has broad IDE distribution.
AI Free3 credits/30 days
AI Pro$100/user/year
AI Ultimate$300/user/year
Enterprisecustom
AI Credit≈ $1
Coding agent in ChatGPT for desktop/mobile/web that can work across codebases.
OpenAI reported Codex users in Business/Enterprise grew 6x since January.
ChatGPT Free/Go/Plus/Pro/Business/Enterpriseincluded by plan
Business annual seat$20/user/mo
codex-mini-latest input$1.50/1M input tokens
codex-mini-latest output$6/1M output tokens
Agentic PR review, code quality, SDLC governance, rules and analytics.
Exact revenue/users not confirmed.
Pro Team$30
Credit pricing$0.012/credit
2,500 credits~18 reviews/mo shown
Enterprisecustom
OS-level long-term memory for developers, local memory, snippets and Copilot.
Pieces claims 1M+ saved materials, 17M+ associated context points and 5M+ copilot messages.
Free$0
Pro$18.99/mo with premium models
Long-term memoryup to 9 months in Pro
Enterprise code search, code intelligence and AI chat/autocomplete over large codebases.
Enterprise devtools vendor; exact current Cody revenue not confirmed.
Enterprisecustom
Old Enterprise Starter$19/user/mo, discontinued
Текущая монетизация проекта
Монетизация выглядит подготовленной, но публично подтверждённой платной версии нет. npm package declares Elastic-2.0 and blocks hosted/managed competing service; README exposes hosted Insight dashboard context-mode.com/insight, but pricing was not found. No confirmed GitHub Sponsors, paid support, paid cloud tier or enterprise pricing.
Коммерческий потенциал
ПОТЕНЦИАЛ · СРЕДНИЙ
Зарабатывать можно на enterprise governance/analytics for AI coding agents. Саму библиотеку продавать трудно; продавать контроль, безопасность, расходы и командную память — нормально.
Спрос и рынок
Спрос сильный. Узкий wedge — context-cost optimization inside AI coding — сидит на рынке AI coding assistants $8B–$10B+ в 2025/2026 и растёт 20–48% CAGR по разным источникам. Реалистичный SAM для independent tooling вокруг agents меньше: enterprise teams с heavy Claude/Cursor/Codex spend.
Ров / защищённость
Есть небольшой technical moat: hooks across 17 clients, routing enforcement, SQLite/FTS5 event store, usage accounting. Нет сильного business moat: platform owners can copy; memory/context vendors crowded; distribution depends on agent clients.
Модели монетизации
- Team/Enterprise dashboard: context savings, model spend, agent productivity, risky tool calls, adoption analytics.
- Policy engine: централизованные routing rules, allow/deny tools, audit logs, SSO/SCIM.
- Private deployment: VPC/on-prem Insight, no-code-retention mode.
- Paid support: rollout для Claude Code/Cursor/Copilot/Codex fleets.
- Security product: sandbox hardening, secrets redaction, MCP risk scoring.
- Marketplace/open-core: free local CLI, paid org analytics and governance.
Что нужно, чтобы сделать продукт
- Подтверждённый paid plan и pricing.
- Enterprise auth: SSO, SCIM, RBAC.
- SOC 2/security docs.
- Deterministic benchmark suite против Cursor/Copilot/Claude Code baseline.
- Admin console для fleet rollout.
- Supply-chain/security audit.
- Понятная лицензия на GitHub с SPDX detection.
- Доказанные customer logos вместо README badges без verification.
⚖ ЛИЦЕНЗИЯ · МОЖНО ЛИ КОММЕРЦИАЛИЗИРОВАТЬ
Не MIT/Apache/BSD. GitHub layer says SPDX None; package/npm says Elastic-2.0. Treat as source-available ELv2 unless LICENSE file proves otherwise. Commercial internal use/support likely possible. Hosted/managed competing service or closed SaaS derivative is restricted. If LICENSE file is absent/mismatched despite package declaration, legal risk becomes high.
Риски и подводные камни
ВЫСОКИЙЛИЦЕНЗИЯ
ELv2 + GitHub SPDX None blocks clean SaaS clone and complicates investment.
ВЫСОКИЙЗАВИСИМОСТЬ ОТ АВТОРА
Claude Code/Cursor/Copilot/Codex can change hooks/MCP behavior and break routing.
ВЫСОКИЙЮР. СЕРАЯ ЗОНА
Tool execution + logs + secrets + MCP require serious enterprise security review.
ВЫСОКИЙКОНКУРЕНЦИЯ
Context7, mem0, Serena, Letta, Graphiti, AgentMemory and commercial IDEs crowd the space.
СРЕДНИЙПРОЧЕЕ
README enterprise badges are not verified by searched sources.
СРЕДНИЙЗАВИСИМОСТЬ ОТ АВТОРА
Owner is individual, verified=False, and project is very young despite ~100 contributors.
СРЕДНИЙКОММОДИТИЗАЦИЯ
Users may churn if native agent context windows or pricing improve.
СРЕДНИЙПРОЧЕЕ
Insight exists, but no confirmed paid SKU/pricing.
Достоверность разбора
УВЕРЕННОСТЬ · СРЕДНЯЯgpt-5.5 · 2026-07-17 · ОКНО 180D
Оговорки / что не проверено
- GitHub enrichment says SPDX None, while README/package.json/npm indicate Elastic-2.0; LICENSE file content was not independently verified here.
- README badges claiming use across Microsoft, Google, Meta, Amazon and others were not verified by independent customer references.
- Context Mode paid pricing, paid cloud tier, paid support, GitHub Sponsors and enterprise SKU were not confirmed.
- npm weekly downloads number was taken from the collected analysis source, not rechecked live.
- Some OSS star counts came from secondary sources/search snippets and ranges, not a fresh GitHub API check.
- rohitg00/agentmemory star count was community-reported and not strongly confirmed.
- riponcm/projectmem stars were not confirmed; mcpmarket showed 0 in collected analysis.
- Commercial scale claims for Windsurf, Augment Code, Amazon Q Developer, JetBrains AI, Qodo and Sourcegraph Cody have missing exact current user/revenue numbers.
- Market-size figures come from analyst/public reports with different methodologies; they are directional, not reconciled.
- Security/sandbox risk assessment is analytical judgment from product surface, not an audited vulnerability finding.
- Moat/capture/access scores are analyst judgment, not measured financial outcomes.
ИСТОЧНИКИ (54)
Why This Is A Finding
mksglu/context-mode собрал 2,678 звёзд за окно, тогда как у автора всего 530 подписчиков — эффективная аудитория ≈ 2,534. Это даёт surprise-индекс 0.00578 (звёзды относительно охвата автора, а не в абсолюте). Удержание форков 93.3% и 158 внешних контрибьюторов отделяют реальный инструмент от разовой вспышки. Акселерация отрицательная — внимание остывает после пика.
Related Findings
RANKS ABOVE 99% OF 2379 FINDINGS
METRICS IN CONTEXT
MEDIAN ACROSS ALL 2379 FINDINGS · Δ vs MEDIAN · PERCENTILE = SHARE RANKED BELOW
METRICVALUEMEDIANΔ MEDPERCENTILE
SCORE0.010.00+0.01ABOVE 99%
VELOCITY14.882.86+12.02ABOVE 93%
RETENTION5.4%5.8%-0.4 PPABOVE 47%
FORKS1,4011,230+171ABOVE 53%
SURPRISE0.010.00+0.01ABOVE 94%