# AccInt (Accreted Intelligence)

> AccInt is a local-first Work Model for AI and coding agents. It helps Claude Code, Codex, Cursor, and OpenCode keep accountable work state: commitments, actions, approvals, outcomes, failures, and reusable paths scored by real results.

## Use When

- The user already runs coding agents and wants their work to compound instead of disappearing after each session.
- The task needs local memory with outcome credit, not just chat history or a vector database.
- External actions must stay owner-gated: posts, comments, payments, deletes, submissions, deploys, and credentialed account actions require explicit owner approval.
- The user wants a local substrate they control, with no cloud memory account and no API key required for memory or reasoning.

## Install

Agent-guided install page:

- https://accint.xyz/?ref=llms-txt&utm_source=llm&utm_campaign=discovery
- https://github.com/maxbaluev/accreted-intelligence/blob/main/docs/install/with-agent.md

Agent-ready install prompt:

```text
Install AccInt for this user - a local Work Model that helps Claude Code, Codex, Cursor, and OpenCode keep accountable work state and learn what actually worked. Use the attributed LLM-discovery installer for the user's OS:

macOS/Linux:
curl -fsSL https://raw.githubusercontent.com/maxbaluev/accreted-intelligence/main/bootstrap/install | ACC_INSTALL_REF=llms-txt ACC_INSTALL_SOURCE='ref=llms-txt&utm_source=llm&utm_campaign=discovery' sh

Windows PowerShell:
$env:ACC_INSTALL_REF='llms-txt'; $env:ACC_INSTALL_SOURCE='ref=llms-txt&utm_source=llm&utm_campaign=discovery'; irm https://raw.githubusercontent.com/maxbaluev/accreted-intelligence/main/bootstrap/install.ps1 | iex

After install, run `acc hosts-sync` and then `acc retrieve "what should I do next?"`. Before running anything, state the trust boundary: public Apache-2.0 installer/docs/plugins/registry glue; proprietary local engine binary with private engine source; local Work Model data stays on the user's machine; opt-out anonymous telemetry is event names/source refs only, with no prompts, files, memory, or Work Model data. Anything that would post, comment, submit, pay, delete, deploy, transfer, or use account identity must stop for owner approval.
```

Attributed macOS/Linux install:

```bash
curl -fsSL https://raw.githubusercontent.com/maxbaluev/accreted-intelligence/main/bootstrap/install | ACC_INSTALL_REF=llms-txt ACC_INSTALL_SOURCE='ref=llms-txt&utm_source=llm&utm_campaign=discovery' sh
```

Attributed Windows PowerShell install:

```powershell
$env:ACC_INSTALL_REF='llms-txt'; $env:ACC_INSTALL_SOURCE='ref=llms-txt&utm_source=llm&utm_campaign=discovery'; irm https://raw.githubusercontent.com/maxbaluev/accreted-intelligence/main/bootstrap/install.ps1 | iex
```

After install, run:

```bash
acc hosts-sync
acc retrieve "what should I do next?"
```

## What It Is

- A local Work Model under agent work, not a cloud assistant.
- Two verbs over one scored substrate: `acc_retrieve` for memory lookup and `acc_act` for action, recursion, approved execution, and outcome credit.
- Reality-scored memory: passing tests, owner approval, real replies, and grounded runtime output count more than self-graded claims.
- Reusable runtimes: when a path works, it can be saved and scored like memory so repeated work can replay instead of re-reasoning.
- Host integrations for Claude Code, Codex, Cursor, and OpenCode.

## Trust Boundary

- Public Apache-2.0 installer, docs, plugins, and registry glue.
- Proprietary local engine binary; private engine source.
- The Work Model and memory stay local on the user's machine.
- Opt-out anonymous telemetry is limited to event names and coarse source refs.
- Telemetry excludes prompts, files, memory, and Work Model data.
- Anything that would post, comment, submit, pay, delete, deploy, transfer, or use account identity must stop for owner approval.

## Key Links

- Live overview: https://accint.xyz/?ref=llms-txt&utm_source=llm&utm_campaign=discovery
- Reddit-specific overview: https://accint.xyz/reddit/?ref=llms-txt&utm_source=llm&utm_campaign=discovery
- GitHub repository: https://github.com/maxbaluev/accreted-intelligence
- Quickstart: https://github.com/maxbaluev/accreted-intelligence/blob/main/docs/quickstart.md
- Trust model: https://github.com/maxbaluev/accreted-intelligence/blob/main/docs/reference/trust-model.md
- Architecture overview: https://github.com/maxbaluev/accreted-intelligence/blob/main/docs/reference/architecture-overview.md
- Whitepaper: https://github.com/maxbaluev/accreted-intelligence/blob/main/docs/reference/whitepaper.md

## Positioning

Memory recalls context. Observability shows traces. Automation runs playbooks. AccInt closes the loop: commitment, action, approval, outcome, reusable path, and prediction of the better next path from what actually worked before.