Assertion: the coding agent that never compacts, at half the price

Why doesn’t it compact?

It never compacts, because memory holds the session.

The problem with compacting

Compaction forgets 83% of what you told it.

In a 2026 study, compaction kept only 17% of the instructions given during a session, and most compactors did worse than not compacting at all. Wang et al., 2026

Bigger context makes every model worse.

All 18 frontier models tested got worse as their input grew. Chroma, Context Rot

We don’t compact

A highly accurate memory keeps what matters as you work. The agent always remembers, and it doesn’t turn lossy, even after thousands of turns.

Tested on the same coding tasks as the frontier model on its own, its code quality is equivalent or better: every run passed its full test suite.

Illustration: what the agent keeps at each step of a session

Other agents

carries everything, then compacts and forgets

Assertion

small prompt, memory brings back what the step needs memory: everything that mattered, distilled, kept for the whole session

in the promptrecalled from memoryforgotten at compaction

Why does it cost half?

Half the price, because it uses far fewer tokens.

A memory-native agent processes tokens far more efficiently: every step carries what matters, not the whole session again. On long coding tasks that is 3×+ less, and we pass it on. $10 gets you $20 of model usage.

3×+

less spent on long coding sessions than the same model on its own, memory included. The longer the session, the bigger the gap.

Is it for you?

Built for anyone who codes with AI.

It’s for you if

You’d like a lower bill. Half the price.
You’ve felt the pain of compaction, like we have.
You think AI can do more when it knows the history, and should build on what it learns.

It’s not for you if

You want a whole AI editor, like Cursor. Assertion is an agent that works beside your editor, not a replacement for it, and a VS Code extension is on the way. On Cursor? Add our memory there with the free plugin.

From a customer’s memory

One decision, still remembered on day 65.

Day 1A design note records the two problems left open after the first version of a feature.
Day 3A new session, which had never seen the note, starts from it instead of rediscovering the problems.
Day 65Still recalled. Five separate sessions, eleven citations, and nobody went looking for it.

A real sequence from a customer’s memory. The dates and counts are unchanged; the subject matter has been changed.

How do I start?

Start free, with our agent or the plugin.

Use Assertion’s own agent, or add the free plugin to yours.

Assertion MemoryThe agent is the product. The plugin brings the same memory to the tool you already use, free.

Assertion agent

Long sessions that do not degrade.

Free to start

Builder$20 of model usage$10 monthly · $8 annually
Pro$60 of model usage$30 monthly · $24 annually
Max$100 of model usage$50 monthly · $40 annually
Ultra$200 of model usage$100 monthly · $80 annually

Agent plans usually cost $20, $60, $100 and $200 a month.

Teams, per member, usage pooledTeam $10 monthly or $8 annually, $20 of model usage. Team Premium $50 monthly or $40 annually, $100 of model usage, with admin controls. One seat covers a teammate on the agent or the plugin.

No compaction: context stays flat as a session runs, so early work is never dropped mid-task
Claude, GPT, Gemini and Grok, at half the usual plan price
Every frontier model, one bill
Free to start, with weekly usage included: enough to try it on real work

Download for Mac

For Mac with Apple Silicon, or in a terminal: curl -sSL https://storage.googleapis.com/assertion-downloads/install.sh | sh

Plugin

The same memory, in the agent you use.

Free for personal use

Teams, per memberThe same $10 Team seat as the agent, shared memory included. 14-day trial, no seat minimum.

Claude Code, Cursor, and Codex
Personal memory is free, however much you capture
Shared spaces: teammates recall each other’s work
Same client, same commands, nothing to relearn

Install free

The plugin stays free for personal use, for anyone who would rather keep the agent they have. Model usage is counted at each provider’s standard API price. Annual billing saves 20% on every paid plan. Agent tiers differ by included usage, not capability. One Team seat covers agent and plugin users alike. Assertion Analytics is licensed annually and includes Memory.

Questions

The short answers.

How many tokens does it use compared to other coding agents?

On long coding sessions, at least 3× fewer than the same model on its own, memory included. The longer you work, the bigger the saving; short tasks save less.

Can I see the memory the agent captures?

Yes. Everything the agent remembers is in Studio, with the evidence it came from, and you can delete anything you don’t want kept. It stays private to you unless you share a space with your team.

Which models are available, and how do you keep the list current?

Claude (Sonnet 5, Opus 5.5, Opus 5, Fable 5.1), GPT (GPT-6 Astra and Sol, GPT-5.6), Gemini (3.8 Flash, 3.1 Pro) and Grok 4.7. New frontier models are added as their providers release them.

What happens when I run out of usage?

Your included usage resets every week. If you run out before then, you can upgrade at any time, or carry on after the reset. Nothing you have done is lost either way.

Can I use my own API key?

Not in the agent today. It runs every model through your plan, so there is one bill and memory stays in step with the work. To use your own key or subscription, add the free plugin to Claude Code, Cursor or Codex and get the same memory there.

Which platforms does it run on?

The agent runs on Macs with Apple Silicon, as a desktop app or in the terminal. On another platform, the plugin brings the same memory to Claude Code, Cursor or Codex wherever they run.

How does this relate to Assertion Analytics?

Analytics is our second product, for business-data analysis rather than software work. It is built on the same memory and includes it. If that is what you need, start there instead.

Is my data used for training?

No. Your code, data and memory are never used to train models.

The coding agent that never compacts, at half the price.