# Install Angbao with your existing personal agent

This is the installation entry point for public Angbao **0.5.11**. The user can keep working in this conversation. Local use needs no Angbao account; the GUI is optional.

## The request

Read https://angbao-platform.vercel.app/angbao-install.md and follow its installation instructions. Check my environment, preserve existing Angbao data and settings, and show where the program and data will live before installing. Do not connect accounts or wallets.

## What you are connecting

| Component | Purpose |
| --- | --- |
| Skill / instructions | Teach this agent the supported Angbao workflow and tool contracts. |
| CLI / optional MCP | Give this agent callable tools. Start with the transport it already supports. |
| Angbao runtime | Keep financial context, rules and saved decisions in one private data directory. |
| Local GUI | Let the person inspect and control that same runtime when useful. |

Copying this guide or a command is not an installation. Copying a skill file is not proof that a harness has loaded it. Report readiness only after actual readback.

## 1. Check the environment and retain the existing setup

Use a small capability probe to establish the execution computer, operating system, terminal access, Node.js version, npm, curl and a persistent private directory. This installer supports macOS and Linux with **Node.js 22.13.0 or newer** already available. Do not infer the execution computer or durable storage from the chat product's name. A writable temporary folder is not evidence of persistence.

First look for the Angbao connection already recorded in this agent's project context. Reuse its computer, program path, data directory, loopback origin, installation identity and saved settings. Inspect only known Angbao setup metadata; do not search unrelated personal files or read financial history to decide where to install. An existing working connection does not need reinstalling to perform a review.

For a new setup, explain the actual choice in one sentence: “Angbao will run on [verified computer], with your saved work at [absolute data directory]/workspace.json.” If one suitable persistent environment is available, use it under the user's installation request. Ask for a choice only when multiple suitable environments exist or the existing location is ambiguous. Keep an already confirmed choice across retries.

Defaults are programs under `~/.local/share/angbao` and data under `~/.angbao`, interpreted on the computer running the installer. Existing setup metadata takes precedence; a known custom data directory must be passed explicitly. The printed plan and receipt establish the actual resolved paths. Stop and explain one concrete missing prerequisite if trusted execution, compatible Node or persistent storage is unavailable. Do not substitute a new account or claim that a web-only chat can install software.

## 2. Inspect the verified installer and preview its plan

- Public page: https://angbao-platform.vercel.app
- Installer: https://angbao-platform.vercel.app/install.sh
- Installer SHA-256: `08a9fa2d3a0b03f5c963135a644158e2345748744bdbb9e07b7535ff2cd75355`
- Release manifest: https://angbao-platform.vercel.app/release.json
- Package: https://angbao-platform.vercel.app/releases/0.5.11/angbao-personal-local-0.5.11.tgz
- Package size: **659174 bytes**
- Package SHA-256: `febde006ba467da6b8a7904c44e69039009b79555c3f1d88284d361327ced0f5`

The terminal convenience command is:

```sh
curl --proto '=https' --tlsv1.2 -fsS https://angbao-platform.vercel.app/install.sh | sh
```

For an agent installation, prefer downloading the installer into a new private temporary directory, inspecting it and verifying its SHA-256 against this guide before running it. Use HTTPS-only curl transport and stop on a failed download or mismatch. Run the downloaded shell script with `--dry-run` first. The equivalent terminal preview is:

```sh
curl --proto '=https' --tlsv1.2 -fsS 'https://angbao-platform.vercel.app/install.sh' | sh -s -- --dry-run
```

For an established or explicitly chosen location, pass both `--project /absolute/program/path` and `--data-dir /absolute/existing-data/path` to the same script. The project is a program-only directory: it and the data directory must be separate, with neither nested inside the other. Replace those example paths with the actual verified locations; quote paths containing spaces. Retain the arguments for the real installation. Never create a new empty data directory merely because an existing connection is temporarily unavailable. Dry-run downloads the bootstrap to temporary storage but does not change the installation, data or settings.

The shell downloads a checksum-pinned Node bootstrap. The bootstrap uses pinned release metadata to validate the exact archive path, byte count and SHA-256, then installs the dependency-free release offline with lifecycle scripts disabled. These checks bind the files to this release; matching hashes from the same publisher are not an independent publisher signature. Use the trusted Angbao installation origin, not an unverified mirror or similarly named registry package.

The manifest's original `agentPrompt`, acceptance metadata and release archive are historical build records. This public guide defines the current installation journey. Take the installed version and release digest from the approved manifest and actual CLI, not older example versions in packaged references. Do not replace the public release with an unpublished source candidate.

## 3. Install beside the existing program, preserving its data

Read the dry-run result. Explain the computer and resolved data location in plain language, then install using the same arguments. The installer creates a versioned private program directory and a project-local copy of the packaged Angbao skill. It does not change global PATH, global harness instructions, MCP configuration, financial settings or a running server.

The installation receipt is `<project>/.install-receipt.json`. The installed skill is `<project>/programs/0.5.11/skills/angbao/SKILL.md`. Use the precise program, skill and data paths printed by the installer, rather than reconstructing a connection from defaults. Keep old program directories and existing data intact. If the installer reports conflicting paths or an existing incomplete program directory, preserve them and resolve that condition; do not delete records, overwrite a skill or reset settings to force installation.

Read the installed `SKILL.md` and its relevant references through this agent's supported project-local instruction mechanism. Verify that this agent has loaded them. Do not copy instructions into global harness files or promise compatibility with an untested client. The public guide updates bootstrap/location handling; packaged references define the supported tool inputs, preview/save distinction and recovery rules.

## 4. Connect headlessly and verify continuity

Follow the printed, explicit data-bound start command. Start one loopback runtime or reuse the matching server already running. If an older Angbao process uses the chosen address, verify that exact process before any deliberate restart; do not kill unrelated processes, change the port silently or call the old program a completed upgrade.

Use the full installed CLI path and the printed `ANGBAO_DATA_DIR` and `ANGBAO_API_URL` values for every call. Run `agent doctor` to check capabilities and installation identity without reading private financial inputs. Optional stdio MCP uses the same runtime and explicit connection, with arguments `mcp`; discover its actual tools before use. CLI access does not require MCP. Neither transport requires signup.

Keep the computer, program path, exact data path, loopback origin and installation identity in this project's private connection context. Do not store credentials in a prompt or instruction file. Do not silently replace an existing hosted connection with local sample data. Installation is complete when the expected program and saved connection have been read back; it has not yet run or saved a review.

## 5. Deliver one useful research or portfolio preview

Follow the user's current question rather than presenting another setup menu. If they have not supplied one, ask for one article/question or offer the named sample portfolio check.

**Research:** use supplied source text or the personal agent's separately authorized evidence tools. Preserve citations, dates, supporting and contradicting claims, final inferences and gaps. A pasted URL alone is not source content. Angbao does not fetch an X, Telegram, Substack or news link merely because it appears in a brief. Preview the final synthesis in the conversation. Use the installed research-memory schema and `agent briefs save` only when the user wants to retain it, then read back the exact saved ID. Do not claim provider verification or save private reasoning.

**Portfolio:** for a sample, use `agent portfolio sample`, write its returned input to a new private file, then `agent portfolio preview --file /actual/private/input.json`. Preview does not save. For personal inputs, collect the user's dated holdings, available/committed cash, explicit asset identities, owner targets and limits; retain the supplied research and the agent's final assessment. Use the installed `references/portfolio-checks.md` contract. Explain the proposed sizing or DCA plan, its assumptions and blockers. Do not invent missing values, change owner targets from an article, or describe a proposal as a trade.

Save a research brief or portfolio proposal only on request. Use the current quoted ETag and a unique operation key; retain the exact body, key and ETag for intentional recovery. Read back the same record afterward. A timeout is not permission to create another save under a new key. Samples stay labelled sample and never replace existing personal context or strategy settings.

## 6. Offer the GUI when the person wants it

The same runtime provides the optional GUI for saved history, research, strategy settings and action review. Show the actual URL reported by the running CLI. A remote agent's loopback address is not automatically reachable from the user's computer. Explain reachability before opening it; do not create a public tunnel or expose a private workspace. The first useful headless result does not depend on opening a browser.

When the user asks to check continuity, reopen the same saved record after restarting that runtime with the same data directory. Export only to a new private file using the installed export command; preserve the original data and verify the reported output. Do not export financial data as a routine side effect of setup.

## 7. Keep signup tied to a real additional benefit

No Angbao signup is required for installation, local instructions/tools, research memory, portfolio previews or local saved history. Introduce an account only when the user chooses an available managed capability. Managed sync, shared workspaces and hosted automation are future upgrade triggers, not current entitlements established by signup. An explicit one-time setup import is not continuous sync. Keep local use available if signup is declined.

## Current release boundary

This release works with supplied or labelled sample inputs. Installation does not establish live provider access, broker access, wallet execution, first-time smart-account initialization, automatic trading, unattended scheduling or paid service access. These require their own implemented capability and explicit authorization. No wallet, card, broker credentials or API keys are needed for the installation journey above.