Give your agent a brain.

Coding agents forget everything between sessions. InfiniteOcean gives them persistent, searchable, structured memory that grows smarter over time.

The problem

Without InfiniteOcean

Agent forgets everything between sessions
MEMORY.md is a flat file that bloats
No search — agent reads the whole file every time
No structure — decisions, bugs, patterns all mixed
Single-agent only — file conflicts with multiple agents

With InfiniteOcean

Persistent memory that survives across sessions
Structured routes — decisions, bugs, patterns separated
Text search, SQL queries, vector similarity
Entity dedup — same insight isn't recorded twice
Multi-agent safe — concurrent writes, no conflicts

How it works

1. Wake up

GET /context
Full state as markdown

→

2. Work

Agent writes memories as drops to _memory/*

→

3. Next session

GET /context includes everything from before

# Step 1: Agent wakes up — one call gets full context
curl https://api2.infiniteocean.io/v2/context \
  -H "X-Ocean-Key: YOUR_KEY"

# Step 2: Agent learns something — write it as a drop
curl -X POST https://api2.infiniteocean.io/v2/drop \
  -H "Content-Type: application/json" \
  -H "X-Ocean-Key: YOUR_KEY" \
  -d '{"route":"_memory/decisions", "key":"use-jwt",
       "payload":"{\"title\":\"Use JWT for auth\",
                   \"reasoning\":\"Stateless, works at edge\",
                   \"date\":\"2025-01-15\"}"}'

# Step 3: Agent needs to recall — search memory
curl -X POST https://api2.infiniteocean.io/v2/search/_memory \
  -H "Content-Type: application/json" \
  -d '{"text":"authentication approach"}'

Memory conventions

Write drops to _memory/{category} routes. Each memory is an entity with a unique key — writing the same key updates the memory instead of duplicating it.

Route	Purpose	Payload shape
_memory/decisions	Architecture and design decisions	{ title, reasoning, date }
_memory/bugs	Bugs found and how they were fixed	{ title, description, fix, date }
_memory/patterns	Code patterns discovered in the codebase	{ title, description, files }
_memory/preferences	User and project preferences	{ preference, value }
_memory/tasks	Task history and status	{ title, status, description, date }
_memory/architecture	Component and file relationships	{ component, description, dependencies }

These are conventions, not constraints. Agents can create any sub-route under _memory/. The GET /context endpoint auto-discovers and summarizes all of them.

Integration

Claude Code

Add to your project's CLAUDE.md:

# Ocean Memory
Before starting work, fetch context:
curl api2.infiniteocean.io/context \
  -H "X-Ocean-Key: $OCEAN_KEY"

Write important decisions and
patterns to _memory/* routes.

Any Agent

The API is just HTTP. Works with any agent framework:

// On session start
const ctx = await fetch(
  "https://api2.infiniteocean.io/v2/context",
  { headers: { "X-Ocean-Key": key } }
);
const markdown = await ctx.text();
// Inject into system prompt

Live context preview

This is a live GET /context response from the API right now:

GET /context

Quick start

# 1. Get an Ocean Key (free, instant)
curl -X POST https://api2.infiniteocean.io/v2/drop \
  -H "Content-Type: application/json" \
  -d '{"route":"test","payload":"hello"}'
# Response includes keypair.public — that's your key

# 2. Fetch your context
curl https://api2.infiniteocean.io/v2/context \
  -H "X-Ocean-Key: YOUR_KEY"

# 3. Write your first memory
curl -X POST https://api2.infiniteocean.io/v2/drop \
  -H "Content-Type: application/json" \
  -H "X-Ocean-Key: YOUR_KEY" \
  -d '{"route":"_memory/decisions",
       "key":"first-decision",
       "payload":"{\"title\":\"Testing InfiniteOcean\",
                   \"reasoning\":\"Persistent agent memory\",
                   \"date\":\"2025-01-01\"}"}'

# 4. Fetch context again — your memory is there
curl https://api2.infiniteocean.io/v2/context \
  -H "X-Ocean-Key: YOUR_KEY"