grepticon/
Sign inGet started

Agents don't need a vector database.
They need a filesystem.

Give your agents retrieval over isolated, permissioned workspaces. No sandbox to boot, no embeddings to maintain.

Start buildingRead the docs
acme-docs — agent session
user › what does our refund policy cover?
→ find "**/*.md"42 paths · 9 ms
→ grep "refund policy"3 matches · 44 ms
policies/returns.md:41  …the full refund policy applies to unopened…
support/faq.md:12  Our refund policy covers 30 days from…
→ cat policies/returns.md2.1 KB · 12 ms
✓ answered from 3 files · 168 tokens read
Works withTyped SDKVercel AI SDKMCP adapterREST /v1grepticon CLI

Simple tools. Zero setup.

Your agents call the unix verbs every model already knows. No query language, no schema, no framework glue.

lslist

List a workspace tree — directories and files, one level at a time.

$ ls acme-docs
policies/  support/  guides/  contracts/
findglob

Match paths by glob pattern across the whole workspace.

$ find "**/*.md"
support/faq.md
policies/returns.md  …41 more
catread

Read a file by path — clean markdown, extracted on ingest.

$ cat policies/returns.md
# Returns & refunds
Items may be returned within 30 days…
grepsearch

Search content across the workspace, ranked, with line numbers.

$ grep "chargeback"
support/faq.md:58  …dispute a chargeback…

Built to get out of the agent's way.

No cold starts

Reads hit a hot, per-workspace index — never a container that has to boot first.

container boot~900ms
grepticon44ms

Least privilege

Mint a token per session. What it can't see doesn't exist to it — invisible, not forbidden.

access tokengrp_at_9f27…a812
group:hrread-only

Nothing off the record

Every read your agents make lands in an append-only audit trail. Export the raw log anytime.

grep"chargeback" · 41ms
catreturns.md · 12ms
find**/*.md · 9ms

Faster and more accurate than RAG.

p95 read latency−6 ms
44ms
Retrieval accuracy
grep-agent.92
agentic-RAG.78
naive-RAG.61
Answer-F1 vs. RAG baselines, corpus v2.
Reads this month+18%
4.2M
For developers

Three lines into your agent loop.

A typed SDK over the /v1 API, a ready-made tool set for the Vercel AI SDK, and a drop-in /mcp endpoint for any client.

acme-docs/
├─ policies/
│  ├─ returns.md        grep "refund" ×3
│  └─ privacy.md
├─ support/
│  └─ faq.md            cat
└─ contracts/
   └─ msa.md            find **/*.md
agent.tsTypeScript
import { GrepticonClient } from '@grepticon/sdk';
import { createVfsTools } from '@grepticon/sdk/ai-sdk';
 
const client = new GrepticonClient({
  apiKey: process.env.GREPTICON_API_KEY,
});
 
// four read tools, ready for your loop
const { text } = await generateText({
  model: anthropic('claude-sonnet-5'),
  tools: createVfsTools(client.session('acme-docs')),
  prompt: 'What does our refund policy cover?',
});
Read the docsBrowse the SDK
Start now

Give your agents a filesystem.

Create a workspace, push your files, and your agents can ls, find, cat, and grep in minutes. Free to start.

Start buildingRead the docs