dartmouth-chat
Skill0 starsChanged 35 days agoBSD-3-ClauseView on GitHub
Help Dartmouth community members use Dartmouth Chat (chat.dartmouth.edu), the Open WebUI-based AI chat platform run by Research Computing & Data (RCD). Use when someone asks about access/eligibility, daily credits and credit groups, choosing models and what the model tags mean, privacy of local vs cloud models, UI features (tools, web search, file uploads, memory, study mode, temporary chat, compaction, image generation, X-Ray), Workspace (custom models, knowledge bases, prompts, skills, automations), or using the Dartmouth Chat API and connecting coding agents (Claude Code, OpenCode, Kilo Code, Mistral Vibe, LangChain).
- Reads credentials
- Installs packages
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- dartmouth-brandingSkill
- decision-modelsSkill
- langchain-dartmouthSkill
- presentation-designSkill
- research-container-deploymentSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 114
Model IDs change: Never present a model list from this file as current.
- Distinctive: line 157
CLOUD models: run under each provider's business/API terms, not consumer terms:
- Distinctive: line 160
OpenAI, Mistral: no training on customer data.
- Distinctive: line 181
Reasoning Effort: use the lightbulb icon by the prompt box, or Chat Controls › Reasoning Effort.
- Distinctive: line 234
Only a limited number of tokens from the collection is sent per message.
- Very specific: line 238
Create: Settings › Automations › New Automation (title, prompt, model, schedule).
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Dartmouth Chat
Dartmouth Chat (https://chat.dartmouth.edu) is a web chat app for large language models, offered by Dartmouth Libraries and Research Computing & Data (RCD). It provides open-weight local models hosted on Dartmouth hardware and cloud models from OpenAI, Anthropic, Google and Mistral. Use is free up to a daily limit.
The canonical docs are https://rc.dartmouth.edu/ai/online-resources/. Each topic below links to its page. Models, versions and menu labels change often. If a detail matters, fetch the linked page and use what it says now over this file.
Ground rules for answering
- Model IDs change. Never present a model list from this file as current. Point users to the model dropdown or
GET https://chat.dartmouth.edu/api/models. IDs have the formprovider.model, e.g.openai.gpt-oss-120b,anthropic.claude-sonnet-4-6. - Cost and privacy follow LOCAL vs CLOUD.
- LOCAL (FREE) models are free and unlimited, and data stays on Dartmouth systems.
- CLOUD ($–$$$) models use daily credits and fall under the provider's business terms.
- Don't promise what isn't published. The docs give no numeric limits for file size, upload count or knowledge-base tokens.
- Support routes:
- Help: research.computing@dartmouth.edu
- Outages: https://status.dartmouth.edu (under "Research Computing")
- What's new: https://dartgo.org/dchat-blog, or Settings › About › "see what's new"
Access
- Eligible: users whose primary affiliation is Faculty, Staff, Student, Emeritus Faculty, or a Sponsored account. Sign in with Dartmouth credentials.
- Not eligible: alumni. There are no individual exceptions. Send alumni to alumni.help@dartmouth.edu.
- Group-limited: some features, such as Workspace, and some usage limits vary by group.
- Sensitive or regulated data: for a more secure LLM service, contact research.computing@dartmouth.edu.
FAQ: https://rc.dartmouth.edu/ai/online-resources/faq
Credits and cost
- Daily allowance: cloud models draw on 2,000 credits per day. The counter is at the upper right. Hover it to see the reset time.
- More
$on a model means it costs more per token and uses the allowance faster. - Out of credits: switch to a FREE/LOCAL model or wait for the reset.
- More
- Hidden costs: reasoning tokens are billed even when the "thinking" isn't shown. Side-by-side model comparison roughly doubles usage. Memory adds tokens on cloud models. Automations can use a lot.
- Estimating cost: the Dartmouth Tokenizer at https://ai-tools.dartmouth.edu/dtok gives free relative-cost estimates.
Credit Groups extra credits for a person, class or program, billed monthly for what's used:
- Request: https://services.dartmouth.edu/TDClient/1806/Portal/Requests/ServiceDet?ID=56401 (group description + members' NetIDs/emails). Members must be Dartmouth community members. External collaborators need a Sponsored Account.
- You receive a Virtual Key.
- In Dartmouth Chat: Profile › Settings › Connections › add a connection with
- URL
https://llm-proxy.dartmouth.edu - Key = the Virtual Key
- Prefix ID = a 3–6 character label for the group
- URL
- Save the connection, then save Settings.
- In the model selector, use the Direct filter to pick the group's models. The group allowance then shows at the upper right.
- For API-only use, use the Virtual Key in place of your personal API key.
Docs: https://rc.dartmouth.edu/ai/online-resources/setting-up-credit-groups
Privacy
- All chats: history and uploads are stored on Dartmouth infrastructure. Data isn't shared between users or used by Dartmouth for training. ITC keeps no logs of prompts beyond what session persistence needs. Data is handled under the policy Access to User-Related Electronic Information.
- LOCAL models: prompts, files and images never leave Dartmouth systems.
- CLOUD models run under each provider's business/API terms, not consumer terms:
- Anthropic: Claude is served via Vertex AI with request/response logging off. Prompts aren't used for training. Exception: Mythos-class models (e.g. "Fable 5") retain prompts and outputs for 30 days for trust and safety.
- Google: prompts may be logged for a limited period for abuse detection, and kept up to 60 days under Google's Advanced AI Safety Addendum.
- OpenAI, Mistral: no training on customer data.
- To keep a chat out of history: turn on Temporary Chat (toggle, top right).
Choosing a model
Open the dropdown at the upper left, or click the tag icon next to a model to see its tags:
| Tag | Meaning |
| FREE / $ – $$$ | FREE = local, no credits. More $ = higher relative cost. |
| LOCAL / CLOUD | LOCAL runs at Dartmouth (private, free). CLOUD runs at the provider. |
| VISION | Accepts images (needed to interpret uploaded images). |
| REASONING | Thinks by default. |
| HYBRID REASONING | Answers directly unless you raise Reasoning Effort. |
| ADAPTIVE REASONING | Picks its own thinking depth (e.g. GPT-5.1 Instant/Codex). |
| TOOL CALLING | Required for any Tools to work. |
Tag definitions: see the tags page.
- Compare models: click
+next to the model name to run two side by side. Settings carry over. - Model health: the Model Status link in the bottom toolbar shows availability. A
?means there's no live status; click it to open the provider's status page. - Reasoning Effort: use the lightbulb icon by the prompt box, or Chat Controls › Reasoning Effort.
- Common values:
high/medium/low, plusminimal/noneon some GPT models. - Click Custom to reset to the default.
- Common values:
- GPT-5 family: these models skip Markdown by default. Add a system prompt in Chat Controls › System Prompt asking for Markdown, and repeat it every 3–5 messages. See https://rc.dartmouth.edu/ai/online-resources/gpt5-notes
Chat features
Most toggles are under the Integration menu below the prompt box.
| Want to… | How |
| Upload documents (RAG) | + › Upload Files. Only VISION models read images. |
| Use specific webpages / YouTube | + › Attach Webpage (YouTube pulls the transcript) |
| Search the web | Integration › Tools › Web Search Tools |
| Generate or edit images | Integration › Tools › Image Tools, then refine with follow-up prompts |
| Make Word/Excel/PowerPoint files | Integration › Tools › Create Document Tool |
| Let the model ask clarifying questions | Integration › Tools › Ask Question Tool |
| Other tools | Knowledge, Chats (query past chats), Notes, Time Utility, Visualization, Email (sends only to your own address) |
| Set default tools | Avatar › Settings › Personalization |
| Learn rather than get answers | Integration › Study Mode. Options icon edits the tutor instructions. |
| Assistant persona | Integration › DChat Persona. Options icon edits it; Default restores it. |
| Remember facts across chats | Integration › Memory (off by default; brain icon shows when on). Manage in Settings › Personalization › Memory. |
| Inspect raw API messages | Integration › X-Ray, then the magnifying glass under a reply |
| Context usage | Hover the ⓘ circle under a reply. Click it for details or Compact Now. |
| Auto-summarize long chats | Settings › Chat Compaction (off by default). Threshold % of the context window, e.g. 75% of 128K = 96K. Not available in temporary chats. |
| Regenerate | Circular-arrows icon: Try Again / Add Details / More Concise |
| Organize chats | Left sidebar. Each chat's … menu: Rename / Share / Clone / Download / Delete. Use tags and folders. |
| Folder defaults | A folder can hold a system prompt and/or knowledge base for new chats in it. Folders can't be shared. |
| Save or share results | … at top right: Share link (recipients can clone), or Download as .json / .txt / .pdf |
| Reopen the welcome tour | User menu › Welcome Guide |
Tool caveats:
- Tools are enabled by the user but triggered by the model. If the model doesn't use one, ask it to explicitly.
- Tools only work on TOOL CALLING models.
- The wrench icon in the message bar shows how many tools are active.
Workspace (build your own chatbot)
Open it from the left sidebar. Every item is Private by default. Models and knowledge bases are shared via Access; prompts and skills via group Read/Write.
- Models: Workspace › Models › New Model. Fields:
- Name, Base Model, Description, Avatar, Tags
- System Prompt, Prompt Suggestions
- Knowledge, Tools, Skills, Filters, Capabilities
- Freeze Model Configuration: turn this on before sharing so others can't change the setup.
- Knowledge: Workspace › Knowledge › New Knowledge.
- Accepts files or a synced folder (Word, PDF, text…). Content must be machine-readable; scanned PDFs often aren't. Review the extracted text.
- Ways to use it:
- Type
#and pick the collection (forces its use). +› Attach Knowledge.- Attach it to a custom model.
- Turn on the Knowledge Tool (the model decides whether to use it).
- Type
- Only a limited number of tokens from the collection is sent per message.
- Prompts: Workspace › Prompts › New Prompt. Supports
{{variable}}placeholders. Run with/. - Skills: Markdown instructions the model loads when needed. Workspace › Skills › New Skill. Run with
$. - Automations: scheduled runs of a custom model + prompt.
- Create: Settings › Automations › New Automation (title, prompt, model, schedule). Supports
{{CURRENT_DATE}}. - Results land in a sidebar folder.
- They use a lot of credits, so schedule them sensibly.
- Create: Settings › Automations › New Automation (title, prompt, model, schedule). Supports
API
- Getting a key: chat.dartmouth.edu › Profile (lower left) › Settings › Account › API Key / Create new secret key. The key shares your account's credit limits.
- Base URL:
https://chat.dartmouth.edu/api - Compatible with: OpenAI Chat Completions, OpenAI Responses and Anthropic Messages clients.
export DCHAT_KEY=... # your API key
# List model IDs
curl -s https://chat.dartmouth.edu/api/models \
-H "Authorization: Bearer $DCHAT_KEY" | jq -r '.data[].id'
# Chat completion (local model = no credits)
curl -s https://chat.dartmouth.edu/api/chat/completions \
-H "Authorization: Bearer $DCHAT_KEY" -H "Content-Type: application/json" \
-d '{"model":"openai.gpt-oss-120b",
"messages":[{"role":"user","content":"What is deep learning?"}],
"stream":false}' | jq -r '.choices[0].message.content'More resources:
- Python:
pip install langchain-dartmouth - Cookbook: https://dartmouth.github.io/langchain-dartmouth-cookbook/00-intro.html
- Tutorial: https://rcweb.dartmouth.edu/~d20964h/2024-12-11-dartmouth-chat-api
Docs: https://rc.dartmouth.edu/ai/online-resources/using-the-api
Connecting coding agents
Full, current configs: https://rc.dartmouth.edu/ai/online-resources/connecting-ai-clients. All of them use the base URL https://chat.dartmouth.edu/api and the user's API key. For best results, pick a Claude or GPT model, with a free local model such as openai.gpt-oss-120b as a fallback.
Step-by-step configs for each tool are in references/coding-agents.md. Load that file when a user is setting one up:
| Tool | Config location | Key points |
| Claude Code | ~/.claude/settings.json | ANTHROPIC_BASE_URL = base URL; key via apiKeyHelper (macOS Keychain) |
| OpenCode | ~/.config/opencode/opencode.json | @ai-sdk/openai-compatible provider; long timeouts |
| Kilo Code (VS Code) | Settings › Providers › Custom provider | ID dartmouth-chat |
| Mistral Vibe | ~/.vibe/config.toml + ~/.vibe/.env | api_style = "openai", backend = "generic" |
Troubleshooting:
- "model not found": the ID must exactly match
/api/models, including theprovider.prefix. - 401: regenerate the key.
- 429 or exhausted credits: switch to a local model or set up a Credit Group.
Beyond chat
- Custom AI applications: RCD's Research Software Engineering team. See https://rc.dartmouth.edu/ai/development-resources/
- Coursework and responsible-use policies (students, Geisel, Guarini, Thayer, Tuck, staff): https://rc.dartmouth.edu/ai/online-resources/policies-bias-responsible-use
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

