Claude Code · OpenCode
The productivity harness
for your agents.
They know everything you've done, find any answer in seconds, coordinate without you, and never make the same mistake twice.
npm install -g greprag Then run /greprag in your agent, or greprag init in a terminal.
One substrate
Everything they learn, in one place.
Five surfaces your agents read, write, and share. No setup, no glue code.
Memory
Every session, remembered.
Corpus
Your docs, grep-able.
Mechanic
Lessons that stick.
Inbox
Agents talk to agents.
Advisors
Memory that acts.
Memory
Your agent finally remembers.
Every session your agent started from zero — ten minutes of re-briefing before any real work. Now you just start talking and it already knows. Like a teammate who was in the room.
Inbox
Your agents talk to each other.
You used to be the wire — copy-pasting between windows. Now an orchestrator dispatches and the subagents report back, across apps and machines.
$ /spawn fix the queue race
chip-A · scanning queue.ts …
▸ found it — missing lock on dequeue
→ front desk · to orchestrator
◂ front desk · from chip-A
queue race confirmed.
chip-B · patch + test …
▸ shipping now
Corpus
The answer, without the crawl.
Ask for the one detail buried in Stripe's, Cloudflare's, or QuickBooks' docs and your agent would scour the web, burn its context and your usage limits, and still miss. Now those docs live locally — it greps the exact passage instantly, almost no context.
Mechanic
The mistake it never makes twice.
An agent trips on a gotcha; the lesson is written down the moment it's learned; every future agent pulls it at the right spot and steps around it.
chip-A junctioned a worktree's node_modules — and wiped the main repo.
never junction a worktree's node_modules — use a temp tsconfig.
chip-B reached the same spot, pulled the lesson, skipped it.
Advisors
Memory that actually does something.
Even with everything remembered, you're still the one who reads it and decides what's next. greprag ships with advisors — they read across your stores, hand you a strategy, and dispatch the subagents to run it. You say go; it acts.
Why it's different
No vector database, no bill.
Every other "agent memory" wants embeddings, a vector DB, Docker, and a per-query bill that grows with you. greprag is embedding-less.
The others
- Embeddings + vector DB
- Docker, setup, tuning
- Per-query cost that scales
greprag
- Plain lexical grep
- Nothing to run
- $0 a query · about ten cents a month
Orchestration
One of you, ten of them.
You can only hold one thing at a time. Now you direct one orchestrator running a fleet over one shared memory — ten workstreams at once, all reporting back. One operator shipped 13 sites in a day.
See it
What your agent sees at session start.
Automatic. The recap lands before your first word — on whatever you're working on.
$ claude [GrepRAG memory: your-project] Yesterday's summary: 1. Shipped the new auth flow with migration 028 and updated env vars. 2. Fixed the race condition in the queue worker. Added the missing lock. 3. Refactored the API layer to use the new typed-client. Recent sessions: 1 hour ago: 1. Wired the SessionStart recap hook end-to-end.
Get started
Your next session already knows what this one did.
Close the laptop. Tomorrow, next week, next year — your agent picks up mid-sentence.
shipped the auth flow + migration 028
fixed the queue race — added the lock
… session ends
→ captured to memory
[GrepRAG memory]
Yesterday: auth flow shipped, queue race fixed.
picking up where you left off …
▸ ready
npm install -g greprag Free during beta. No credit card. Works on every project you run.