Autonomous
The persistent knowledge and control layer: recall, NAV-routed reading, continuity, extraction, ownership, and evidence-aware operations through a local MCP interface.
Autonomous CPC
Persistent knowledge. Local control.
Autonomous is CPC's local persistent knowledge and control engine. It gives supported AI clients a durable way to recall context, navigate private knowledge, preserve continuity, and carry out governed operations on your own machine.
Autonomous is not a public download yet. AutoCache is its customer site, where the current preview and the support, privacy, and terms pages live.
autonomous + cpc
client> recall the live project state and continue safely
navigate local knowledge
recall routes through catalogs and source files
continuity restores owned operations and evidence
control applies consent and ownership boundaries
verification separates installed from proven
local-first, persistent, user-controlled
Private distribution
One engine, with explicit layers.
The package is designed to make the required procedure layer clear, keep optional behavior optional, and give a new installation a privacy-safe place to start.
autonomous-skills
The required CPC procedure skills and Autonomous MCP registration. This is the base profile for every supported installation mode.
autonomous-hooks
Scoped Codex lifecycle and audit hooks. They supplement the base skills; they do not replace them and are not enforcement merely because they are installed.
Starter Volumes
A synthetic pack of 9 physical Volumes representing 10 logical units, demonstrating the structure without copying a real person's documents, history, credentials, or operational state.
Opt-in scheduling
Previewable upkeep tasks for a receiving Windows computer. Nothing is registered or enabled automatically; registration requires a reviewed preview and an explicit user action.
Installation flow
Install the source package, then prove each activation stage.
Repository access starts the process. Installation is not the same thing as a ready runtime, an invoked hook, or proven enforcement.
Start at AutoCache.Autonomous is distributed through its customer site rather than a public repository. Request the current preview at autocache.ai, then use the verified package you receive on the receiving Windows computer.
Validate before installation.Run the included validator and review its source-integrity, dependency, and runtime-boundary results.
Install the required base.Install autonomous-skills. Add autonomous-hooks only when the receiving Codex host needs the optional hook profile.
Start with synthetic knowledge.Use the Starter Volumes pack as a privacy-safe baseline, then add only the user's own reviewed material.
Prove activation separately.Confirm source found, installed, loader discoverable, real-event invocation, and enforcement status as separate states.
Opt in to upkeep only after review.Preview every proposed Windows task. Register and enable it only through a separate, explicit user action.
Trust boundaries
What the package does not claim.
Autonomous is intentionally explicit about the gap between source, installation, discovery, invocation, and enforcement.
No user data bundled
The distribution does not include a user's Volumes, credentials, configuration, logs, transcripts, backups, identities, or live operational state.
No executable currently bundled
The Autonomous runtime is a separate release concern. Runtime readiness remains separately proven through an approved build, expected hash, and local protocol checks.
Hooks are not enforcement until trusted and real-event tested
Hook source, installation, and discovery are not enough. Exact definitions must be reviewed and trusted, then exercised through a route-complete real event on the receiving host.
Scheduled upkeep is never auto-enabled
Windows upkeep requires a preview, explicit registration, and a separate enable decision. The package does not silently create or activate scheduled tasks.
A successful install proves only that installation step. Runtime readiness, hook invocation, hook enforcement, and scheduled-task activation each require their own evidence.
Optional companion
Add interface control only when you need it.
Autonomous is the knowledge and control engine. AI-Hands is a separate optional MCP companion for browser, Windows UI, screenshots, vision, and OCR.
AI-Hands
Use AI-Hands when an authorized agent needs to operate and inspect a graphical interface. It complements Autonomous; it is not required for the Autonomous knowledge layer.
Open AI-Hands repository