HypAware

HypAware is the improvement loop for all your agents. It records every session from your AI clients (Claude Code, Codex, Cursor, OpenCode, and more) into one queryable history on your machine. Then it helps you find what went wrong, what keeps repeating, and what to fix.

• Collect. A lightweight background daemon captures sessions from the clients you already use. No changes to how you work.
• Store. Sessions land in a local cache of open table files, linked as a graph of sessions, repos, files, tools, and skills. No data warehouse.
• Analyze. Query everything with SQL or search, or have your client answer a question in plain English.
• Act. HypAware turns patterns in your history into concrete fixes, like a skill worth adding, with the sessions that prove it.

Install

Requires Node.js 22.12 or newer, on macOS or Linux.

npm install -g hypaware
hyp setup

You can also ask your AI agent to install HypAware using these same two commands. Without a terminal, hyp setup prints a guide so the agent can walk you through the choices and run unattended setup. You complete any browser sign-ins yourself. Agents using a terminal can read the guide with hyp setup --guide.

On a terminal, setup first asks how to collect. Sync to the cloud is the default: press Enter and a browser sign-in enrolls this machine, so your history follows you across machines. Choose Local only to keep recordings on this machine. Then it asks which clients to capture, installs the background daemon, and starts recording. It ends with a first look at your history: tokens per model, activity per day, which repos you worked in, and which tools got called. Both hyp and hypaware run the same CLI.

What you can do with it

See the summary any time:

hyp query overview

Have your client answer a question about your history. hyp ask opens one of the clients HypAware is recording, such as Claude Code or Codex, on the question:

hyp ask "which sessions touched the auth module"
hyp ask            # suggest a skill based on your recent sessions

Or ask from inside any session. Install the HypAware skills and your client looks up past sessions itself whenever a question calls for it:

hyp client skills install

Search and query directly, without an AI client:

hyp query grep "connection refused"
hyp query sql "select count(*) from ai_gateway_messages"

See queries for more.

Reports

Reports review your AI work over a period: token usage, recurring tasks, tool failures, and opportunities to improve. Your agent investigates recorded sessions and proposes specific fixes with evidence:

hyp report generate "Cover last week and focus on repeated debugging work"

We recommend Fable or Astra for the best report quality.

Read the report locally or publish it to make it appear in your team's HypAware Cloud dashboard. See reports for generation and publishing.

Supported clients

Choose Ollama in hyp setup, then run hyp ollama setup for the live CLI/SDK capture host. Run hyp ollama verify --model <installed-model> to confirm a fresh saved request and response. See Ollama setup, recording and direct recovery for supported versions and privacy limits.

Claude Code, Claude Desktop, Codex (CLI and Desktop), Cursor, OpenCode, OpenClaw, Hermes Agent, Pi, Factory Droid, Ollama, and any tool that exports OpenTelemetry logs, traces, or metrics.

NanoClaw v2 Claude transcript capture is supported with an explicitly configured local installation root.

Claude Code is captured through its built-in telemetry, so it still talks directly to Anthropic and nothing sits in the path of your session. See clients and history, including how to import sessions you ran before installing.

Use it with your team

Choosing Sync to the cloud in hyp setup signs the machine in. To sign in a machine you set up as local only, run:

hyp remote login

Everyone's history then flows to HypAware Cloud, so you can analyze usage, spend, and failure patterns across the whole team:

hyp query sql "select count(*) from ai_gateway_messages" --remote

First-time browser enrollment provides a review period when its hold is saved. Setup then asks Upload now? [Y/n]: pressing Enter chooses Yes and uploads immediately, so answer n to keep the review period. Check the deadline with hyp status. See the first-sync review before sharing recordings.

One person can sign in and sync on their own. To put more than one person in an organization, contact us and we'll set it up; there is no self-serve invite yet. See the team setup guide.

Privacy controls

Recordings stay local until you configure Cloud sync or another export destination. You control what is recorded, per folder:

hyp privacy ignore              # never record sessions in this repo
hyp privacy set . local-only    # record, but never send to HypAware Cloud
hyp session ignore              # stop recording the current session
hyp privacy purge --session ID  # delete what was already recorded

See what HypAware records and how to control it.

Is it working?

hyp status

This shows the daemon, your clients, and anything that needs attention. See troubleshooting.

Uninstall

hyp leave                  # only if you signed in to HypAware Cloud
hyp daemon uninstall       # stop the daemon and detach every client
npm uninstall -g hypaware
rm -rf ~/.hyp              # delete all local recordings

Your clients' settings are restored on the way out. Copies already sent to HypAware Cloud or exported to files are not affected.

Documentation

Start at the documentation index for setup, clients, queries and reports, Cloud and teams, privacy, configuration, and troubleshooting. The CLI reference lists complete command syntax and flags.