Built on Gemini & Google Cloud

Give your AI agents a memory that survives the session.

ContextForge is a managed persistent-memory layer for AI agents. Store everything intelligently. Retrieve only what matters. Remember more. Send less context.

Give your agent a memory Read the docs

Free plan available · No credit card required · Integrate in ~5 minutes

The problem

Agents forget everything between sessions — or developers resend enormous histories.

Every AI product team ends up rebuilding the same infrastructure: a way to remember users, decisions, and preferences across sessions. Without it, agents either forget everything the moment a session ends, or developers stuff entire conversation histories into every prompt — burning tokens, latency, and money on context the model doesn't need.

Without persistent memory

❌ Full conversation history re-sent on every request

❌ No way to know what's actually relevant to the current task

❌ Preferences silently go stale or contradict each other

❌ Every team reinvents extraction, storage, and retrieval from scratch

The solution

ContextForge intelligently stores, consolidates, and retrieves durable agent memory.

Decides what matters

Gemini extracts atomic, durable facts and discards filler.

Stores it efficiently

Structured, typed, versioned memories with provenance.

Resolves conflicts

Detects contradictions and prefers the newer truth.

Retrieves only what's relevant

Semantic search + Gemini reranking, token-budget aware.

How it works

Remember → Understand → Retrieve → Act

Remember

Send agent/user content to ContextForge as it happens.

Understand

Gemini extracts, classifies, scores, and consolidates memories.

Retrieve

Ask for context for a new task; get back only what's relevant.

Act

Your agent responds with full continuity, at a fraction of the tokens.

Developer experience

Integrate in about 5 minutes.

A clean REST API, a minimal Python SDK, and explainable responses at every step — so you always know what was stored, why, and what was retrieved.

View full API reference

import contextforge

# 1. Your agent learns something about a user
contextforge.remember(
    user_id="user_123",
    content="I prefer vegetarian restaurants and usually eat around 7 PM.",
)

# 2. Later -- a new session, a new task
context = contextforge.get_context(
    user_id="user_123",
    query="Recommend dinner options tonight.",
)

print(context.compact_context)
# - [preference] User prefers vegetarian restaurants.
# - [preference] User usually eats around 7 PM.

Token economics

Stop paying to re-send history the model doesn't need.

ContextForge estimates exactly how many tokens you avoid sending, project-wide and per retrieval, so the ROI of switching to managed memory is visible from day one.

Context tokens avoided18,420

Context reduction83%

Example from ContextForge's demo project. Your numbers are calculated live from your own stored and retrieved memory content.

Use cases

Built for AI-native teams of any size

AI assistants

Remember preferences, ongoing projects, and how a user likes to work.

Customer support agents

Recall prior tickets, resolutions, and account context instantly.

Sales agents

Track deal history, stakeholder preferences, and next steps across calls.

Developer agents

Persist project conventions, decisions, and past debugging context.

Founder / ops agents

Keep track of decisions, hiring context, and investor conversations.

Pricing

Simple plans that scale with you

Free

Try ContextForge on a single project.

$0/mo

500 memories

1,000 API calls / month

Playground access

Community support

Get started

Give your agent a memory.

Start free, store your first memory in minutes, and see exactly how much context you stop wasting.

Create your free account