Run reviews and save research with your existing agent, without an Angbao account.
Get the current Angbao installer and start guide ↗
Your agent first checks where it can run commands and keep files. Keep with this agent is the recommended choice when that storage is available. The agent should name the computer and folder: a cloud agent’s files may stay on its cloud computer.
On a shared computer, agents using the same system account may be able to read this folder. Your agent should explain that access before you choose.
Keep on my computer is available when your agent has a working connection to it. Confirm the location before transferring or installing Angbao. Keep using an existing installation in the place you already chose.
Your agent checks what this computer needs and follows the private installation guide. The setup file comes from the person providing this Angbao release.
Your agent remembers the location you confirmed and continues from the last verified step after an interruption. Your saved figures, review settings and earlier results stay in place.
If a file is missing or the computer is asleep, your agent explains the one step needed to continue. You do not need to choose your storage again.
Terminal access, Node 22.13 or later and storage that survives restart are required. The supplied guide is CLI-DISTRIBUTION.md, also included as references/connection.md with the skill. The resumable controller checks the supplied runtime and archive, records progress privately, and reuses the verified workspace. New programs install separately; an old server must be stopped by the agent only after its identity is verified. The manual fallback is only for an owned empty project. It cannot wake a computer or download a runtime. It does not change global agent settings, credentials or PATH. Run one server for the recorded data folder.
Choose Try a savings review to preview a complete example before saving it. There are no internal settings to fill in. The example does not replace your own figures or review settings.
For your own review, supply your figures and check the approach before running. No signup is required.
Open a second terminal on the same computer. Use the installed binary directly; no global command-path change is needed. Replace the project path with your chosen location.
/chosen/angbao-project/programs/0.5.9/bin/angbao agent sample preview --preset try-savings
# Save only when requested, with the current revision from that preview.
/chosen/angbao-project/programs/0.5.9/bin/angbao agent sample save --preset try-savings --etag '"N"' --key sample-first-001No Angbao sign-in is needed. To review your own values, choose Start first review in the workspace, or save a complete private declaration with /chosen/angbao-project/programs/0.5.9/bin/angbao agent setup --file /chosen/context.json --etag '"0"' --key local-setup-001. Use the current revision returned by Context. Sample values remain labelled sample.
Resume the recorded setup when restarting Angbao. Your settings, reviews and research stay in the same saved workspace. If you start on a different port, set ANGBAO_API_URL to the address shown before running agent commands.
Each review saves the strategy version, inputs and settings used. To save through the agent, send the current quoted revision and a unique operation key. If the response is uncertain, keep the original key, request body and revision for a deliberate retry. A new key would create a separate review.
For stdio MCP, use the installed binary’s absolute path with arguments ["mcp"]. Start the local server first and choose one connection method. This bridge uses protocol 2025-06-18; each external client still needs a connection test.
Open a review of your own figures and choose Review again. Confirm the current inputs or update them before running. Angbao keeps the original strategy version and settings, then shows what changed between the saved reviews.
From your agent, use /chosen/angbao-project/programs/0.5.9/bin/angbao agent repeat --review SAVED_REVIEW_UUID --inputs-confirmed yes --etag '"2"' --key local-repeat-001 with the current saved revision. A due date is a prompt to review; it does not schedule a background job.
Save and revisit your agent’s final research. Open Research to write a brief with the guided form, preview it and save. Advanced JSON remains available for imports.
Save the final answer, sources and remaining questions. Briefs stay unverified. Saving does not call Elfa or a model. Leave out credentials and private reasoning.
For an optional local portfolio check, choose Research → Check against portfolio. Try the labelled example or enter your own values and targets, add your agent’s assessment, preview and save. Check again preserves the selected targets. A proposed split does not schedule trades or reserve money.
When you want to use this setup in an account, choose Agents → Export setup. Sign in to the account workspace and choose Agents → Import setup. Review the account and settings before importing.
You can keep working locally if you stop or cannot sign in. Keep your local copy until the import is confirmed. This is a one-time import; automatic sync is not available.
Credentials, access keys, plans and financial permissions do not import. Optional agent access to an account uses an owner-issued key with selected scopes and expiry. Keep ANGBAO_AGENT_TOKEN in the agent’s secret environment and set ANGBAO_REMOTE_ORIGIN to the exact HTTPS origin.
Choose account mode explicitly with /chosen/angbao-project/programs/0.5.9/bin/angbao agent --hosted or /chosen/angbao-project/programs/0.5.9/bin/angbao mcp --hosted. Environment credentials alone do not change the default local mode. Never paste access tokens in chat, arguments, URLs or exports. Failed account access does not switch to sample data.
Local terminal use has been tested. Account sign-in, email delivery and individual MCP clients still need end-to-end testing. Future work includes linked workspaces, collaboration, venue connections and Robinhood Chain controls. Installing Angbao does not allow an agent to trade, sign transactions or spend money.