Build with your coding agent

Add Oort memory to this project. Oort is open-source long-term memory for AI agents: the oortmem 0.9.0 package on npm (npm i oortmem, Node.js 22.15 or newer).

First read https://oortcloud.dev/llms-full.txt, or load the oort-sdk skill if it is installed. Use only the names it documents: do not guess a method, option or flag.

Then build it the Oort way:
1. Ask me which mode to use, unless this project already says: local (memory on this machine) or cloud (Oort Cloud, with the key in OORT_API_KEY).
2. Create one handle per process with createOort(), and close it (or use `await using`). In server code pass the options explicitly: mode, home (local), org, project, projectConfig: false and keySources.
3. Before each model call, recall and put formatForPrompt(result) in the system prompt. After answering, observe the turn, and await it before a serverless function returns. Use remember only for facts the user stated or confirmed.
4. Memory reaches a prompt only through formatForPrompt. Never concatenate memory text into a prompt.
5. Keys come from the environment or a secret manager, never from code, a committed config, a prompt or a log line. On a server pass keySources: ['env'].
6. If the app has end users, keep one root handle per tenant (org, never 'local') in trusted server code and give the model only oort.scoped({ user: userId }). A missing user id throws user-missing: do not catch it and fall back. In cloud mode, end users need a server key ("Acts for end users") and oortmem 0.8.0 or newer; with an older oortmem, use local mode.
7. Model-facing code runs in the agent judge role. Use judge: { role: 'expert' } only on a code path where a person really decided.
8. Handle errors by OortError.code. Retry only what is retryable, waiting retryAfter seconds when it is set.
9. Do not send secrets to memory.

When you are done, go through the skill's "Before you finish" list.

Add Oort memory to this project. Oort is open-source long-term memory for AI agents: the oortmem 0.9.0 package on PyPI (pip install oortmem, Python 3.10 or newer).

First read https://oortcloud.dev/llms-full.txt, or load the oort-sdk skill if it is installed. Use the native client (from oortmem import Oort, or AsyncOort) and only the names the docs give: do not guess a method, option or flag.

Then build it the Oort way:
1. Ask me which mode to use, unless this project already says: local (memory on this machine; needs Node.js 22.15 and the oort CLI) or cloud (Oort Cloud, with the key in OORT_API_KEY; no Node).
2. Create one handle per process with Oort(...), and close it (or use `with Oort(...) as oort:`). In server code pass the options explicitly: mode, org (local), project and project_config=False.
3. Before each model call, recall and put format_for_prompt(result) in the system prompt. After answering, observe the turn. Use remember only for facts the user stated or confirmed.
4. Memory reaches a prompt only through format_for_prompt. Never concatenate memory text into a prompt.
5. Keys come from the environment or a secret manager, never from code, a committed config, a prompt or a log line. On a server set OORT_NO_KEYCHAIN=1.
6. If the app has end users, keep the root handle in trusted server code and give the model only oort.scoped(user=user_id). A missing user id is refused: do not fall back. In cloud mode, end users need a server key ("Acts for end users") and oortmem 0.8.0 or newer; with an older oortmem, use local or remote mode.
7. Model-facing code runs in the agent judge role. Use judge_role="expert" only where a person really decided.
8. Handle errors by OortError.code. Retry only what is retryable, waiting retry_after seconds when it is set.
9. Do not send secrets to memory.

When you are done, go through the oort-sdk skill's "Before you finish" list.