- Category
- AI Tools
- Rank
- No. 2099Tools index
Previous survey · No. 2094 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- thedotmack
- GitHub
- 7 stars
- Date
About
Domain-flexible temporal intelligence engine that applies to any document corpus for time-aware retrieval and reasoning.
What it does
Rad Mem watches Claude Code sessions, records prompts and tool activity, then uses the Claude Agent SDK to produce structured observations and progress summaries. It stores that memory locally, restores recent context in later sessions, and supports natural-language searches across decisions, fixes, files, concepts, prompts, and timelines.
Why it's ranked here
The design addresses a concrete agent workflow: retaining useful project history without loading every old transcript into each session. Layered results expose compact indexes before detailed records, while citations connect answers to stored observations. Its value is strongest inside Claude Code, and its broader protocol API appears less documented than that primary integration.
What's good
Capture runs through lifecycle hooks, so ordinary coding work creates memory without manual note taking. Search combines SQLite full-text indexes with Chroma semantic retrieval. Results can be filtered by project, type, concept, file, or time range. The parser preserves incomplete observations and summaries instead of silently discarding potentially useful session history.
Tradeoffs
The main workflow depends on Claude Code, the Claude Agent SDK, a local Node service, PM2, SQLite, and Chroma. That is substantial machinery for session notes. Generated records may contain missing fields because the parser deliberately retains incomplete output. Endless Mode remains experimental and reportedly adds 60 to 90 seconds of latency for each tool observation.
How to use it well
Use it for long-running Claude Code projects where decisions, bug fixes, and implementation context regularly cross session boundaries. Start searches with compact index results, then request full records only for promising matches. Keep source control and original transcripts as authoritative records. Rad Mem organizes recall, but it does not replace those deeper evidence layers.
Technical notes+
package.json defines an ESM TypeScript and Node package requiring Node 18 or newer, with build and Node test scripts plus dependencies on Express, PM2, better-sqlite3, the Claude Agent SDK, and the Model Context Protocol SDK. src/services/worker-service.ts implements the local Express worker, Claude Code session routes, platform-agnostic RAD Protocol routes, retrieval endpoints, server-sent events, settings, and search proxies. src/servers/search-server.ts exposes MCP search backed by SessionSearch and optional Chroma queries, formatting durable rad-mem:// citations. src/sdk/parser.ts parses XML observations and summaries, defaults missing or invalid observation types to change, and retains nullable fields. src/shared/paths.ts places the database, vector store, archives, logs, backups, and settings beneath ~/.rad-mem by default. src/shared/worker-utils.ts uses RAD_MEM_PORT with 38888 as the default and starts the worker through PM2. src/sdk/parser.test.ts contains regression coverage for incomplete observations, incomplete summaries, invalid types, whitespace, and explicit summary skipping.
Observed
- License
- AGPL-3.0
- Primary implementation
- TypeScript targeting Node.js 18 or newer, packaged as ECMAScript modules
- Installation surface
- Claude Code marketplace plugin installed with the built-in plugin commands
- Interfaces
- Claude Code hooks and skill, local web viewer, HTTP API, server-sent events, and MCP search server
- Storage and search
- Local SQLite storage with FTS5 full-text search plus Chroma vector search
- Process model
- Local Express worker managed by bundled PM2
- Testing
- Node test script and parser regression tests are present
Read from README.md, package.json, src/sdk/parser.ts, src/sdk/prompts.ts, src/shared/paths.ts, src/utils/logger.ts, src/sdk/parser.test.ts, src/types/transcript.ts, src/utils/silent-debug.ts, src/shared/worker-utils.ts, src/servers/search-server.ts, src/services/worker-types.ts, src/bin/cleanup-duplicates.ts, src/services/worker-service.ts.
What it can do
Retrieve documents based on temporal queries
Time-based search query and document corpus → Relevant documents filtered by time criteria
Analyze temporal patterns in document collections
Document corpus with timestamps → Timeline analysis and temporal insights
Perform time-aware reasoning across documents
Document corpus and reasoning query with temporal context → Contextual answers considering time relationships
Index documents with temporal metadata
Document corpus → Time-indexed searchable database
Extract chronological sequences from text
Text documents → Ordered timeline of events and information
Compare document states across different time periods
Document versions with timestamps → Temporal comparison and change analysis
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.