AI coding agents (Claude Code, Codex, etc.) generate full session transcripts but there's no good way to search across them or recall past context
Index all agent session transcripts into a searchable database, expose via MCP server or API so agents and users can query their own conversation history across sessions
Freemium - free for local/individual use, paid tier for team-shared memory, cloud sync, and advanced semantic search
The pain is real but intermittent. Developers feel it acutely when they NEED to recall a past session ('how did I fix that auth bug last week?') but forget about it the rest of the time. The pain signals from the GitHub thread are genuine but the 57-star engagement suggests it's a recognized annoyance, not a hair-on-fire problem yet. As agent usage deepens and session counts grow into the hundreds, this pain will compound.
TAM is constrained to developers actively using AI coding agents — roughly 5-10M developers today, growing fast. But the addressable slice who'd pay for this is smaller: power users generating 10+ sessions/week who actually need to look back. Realistic SAM is maybe 500K-1M users. At $10/mo paid conversion of 2-5%, that's $1.2M-$6M ARR ceiling for an indie product. Not a unicorn, but a solid indie/small-team business.
Developers already pay $20/mo for Claude Pro, $20/mo for Cursor, $10/mo for Copilot. Budget fatigue is real. A search-over-sessions tool is a 'nice to have' utility, not a core workflow tool. Free/open-source alternatives will always compete. Best path to payment is team features (shared memory, onboarding context) where companies pay, not individuals. Individual WTP is likely $5-10/mo max.
Highly buildable. Claude Code stores sessions as JSON in ~/.claude/sessions. Cursor and others have similar local storage. MVP is: parse session files → index into SQLite with FTS5 or a local vector DB → expose via MCP server. A competent solo dev can build a working local-only MVP in 2-3 weeks. Semantic search adds complexity but local embedding models (e.g., sentence-transformers) are mature. No novel AI research required.
The gap is wide and clear. Nobody does cross-tool agent session search today. Mem0 and Zep store extracted facts, not full transcripts. Native platform search is siloed and shallow. MCP memory servers store snippets. The specific value prop — 'search across ALL your agent conversations like you search code' — is completely unserved. First mover has a real window.
Local/free tier keeps users in ecosystem. Cloud sync, team sharing, advanced semantic search, and cross-device access are natural paid tiers. Recurring value comes from the index growing over time — the more sessions indexed, the more valuable it becomes. Lock-in is moderate (your search index). Risk: if Claude Code or Cursor build native cross-session search, the paid tier collapses.
- +Clear, unserved gap — no product does cross-tool agent session search today
- +Technically simple MVP with high signal-to-effort ratio
- +MCP ecosystem is the perfect distribution channel — agents can self-discover and use it
- +Growing pain that compounds as agent usage increases and session counts pile up
- +Open source first builds trust with developer audience and drives organic adoption
- !Platform risk: Claude Code, Cursor, or Copilot could ship native session search and kill the market overnight
- !Session format fragility: each tool stores sessions differently, formats may change without notice, requiring constant maintenance
- !Willingness to pay is uncertain — this may be a 'should be free' utility in developers' minds
- !Privacy/security sensitivity: indexing all agent conversations touches proprietary code, making enterprise adoption harder
- !Small engaged market today — 57 stars suggests demand is early-stage, not proven at scale
Memory layer for AI apps that extracts 'facts' from conversations and stores them as structured memory entries with semantic search via API and MCP server
Long-term memory service for AI assistants with automatic summarization, entity extraction, temporal awareness, and knowledge graphs
Framework for building stateful AI agents with persistent self-managed memory across sessions using tiered memory
MCP-based memory server that gives Claude and other MCP clients persistent key-value style memory across sessions, stored locally
Built-in conversation history with basic title and keyword search within each platform's web UI
Local CLI tool + MCP server that indexes Claude Code session transcripts (~/.claude/sessions/) into SQLite with FTS5 full-text search. Expose two MCP tools: 'search_sessions' (keyword/semantic query) and 'get_session_context' (retrieve specific session details). Ship as a single pip/npm install. Week 1: parser + indexer. Week 2: MCP server + basic search. Week 3: add Cursor session support. Week 4: polish, README, launch on HN and r/ClaudeAI.
Free local-only tool (open source, unlimited sessions) → Paid individual tier ($8/mo: cloud sync across machines, semantic search with embeddings, auto-tagging) → Team tier ($15/user/mo: shared team memory, onboarding context from past sessions, access controls) → Enterprise (SSO, audit logs, on-prem deployment, compliance features)
8-12 weeks. Weeks 1-4 for MVP build and launch. Weeks 5-8 for open-source traction and iteration based on feedback. Weeks 8-12 to ship cloud sync paid tier and convert early power users. First dollar likely month 3, meaningful MRR ($1K+) by month 6 if adoption catches.
- “there's no good way to search across them”
- “persist full session transcripts but no good way to search”
- “I don't want a blanket memory”
- “I really want to get them out of main context asap”