Travel Policy 2026Flights
3.1 Class of travel
Book economy for any flight under six hours. Longer flights may use premium economy.
3
Keep your OpenAI client. Change the base URL. Answers cite the page.
Ask questions about a project from the OpenAI SDK or any client that speaks Chat Completions. Set the base URL to DocSlurp, use a DocSlurp API key with the chat scope, and pick a project as the model. Answers come from that project's documents and list their sources.
chat.tsstream: true
What changed in the 2026 travel policy?
Economy class is now required on flights under six hours.[1] The nightly hotel cap rose to $240 in major cities.[2]
Sources:
[1] travel-policy-2026.pdf, page 3
[2] travel-policy-2026.pdf, page 7
Travel Policy 2026Flights
3.1 Class of travel
Book economy for any flight under six hours. Longer flights may use premium economy.
3
Travel Policy 2026Lodging
7.2 Hotel limits
The nightly cap is $240 in major cities and $160 elsewhere.
7
See how RAG builds answers from evidence and what a citation can and can’t prove.
In project chat, source cards show the cited passage and its page image when a rendered image is available. Select an image to open the page reader at that citation. Use the arrow keys to turn pages, Home or End to jump to the first or last page, and Escape to return to the conversation.
Some documents have extracted text without page images. Their source cards still show the passage and document link. If a citation does not identify a page, DocSlurp keeps the excerpt instead of guessing which page to show.
In project chat, click the microphone, speak, then stop. Review the transcript in your message draft and send it to get a cited answer from your documents. You can cancel a recording or keep typing at any time.
Voice input needs microphone permission and HTTPS or localhost. Recordings stop after two minutes. Transcription runs on a hosted speech model that DocSlurp manages; your recording is sent for transcription and is not saved as an uploaded document. Answers appear as text.
Wait for processing to finish in your project.
Give it the chat scope on the API keys page. New server keys have it by default.
Use project:default, project:<slug>, or a model ID from /v1/models.
Create a key at API keys and keep it on your server.
List your projects, then point the client's baseURL at DocSlurp and use a project as the model.
curl "https://docslurp.io/v1/models" \
-H "Authorization: Bearer $DOCSLURP_API_KEY"import OpenAI from 'openai';
const client = new OpenAI({
apiKey: process.env.DOCSLURP_API_KEY,
baseURL: 'https://docslurp.io/v1',
});
// Your organization's default project. For another project, use
// project:<slug>, or a project:<uuid> ID from client.models.list().
const model = 'project:default';
const messages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = [
{ role: 'user', content: 'What are the key findings?' },
];
const answer = await client.chat.completions.create({ model, messages });
console.log(answer.choices[0].message.content);
// Send the history with your next question.
messages.push({ role: 'assistant', content: answer.choices[0].message.content ?? '' });
messages.push({ role: 'user', content: 'Which source supports that?' });
const stream = await client.chat.completions.create({
model, messages, stream: true,
stream_options: { include_usage: true },
});
for await (const chunk of stream) {
process.stdout.write(chunk.choices[0]?.delta.content ?? '');
if (chunk.usage) console.log(chunk.usage);
}[DONE].Messageanswer.choices[0].message.content
curl -X POST "https://docslurp.io/v1/chat/completions" \
-H "Authorization: Bearer $DOCSLURP_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "project:default",
"messages": [{"role":"user","content":"What are the key findings?"}]
}'Send text messages with system, developer, user, or assistant roles: up to 64 messages and 128,000 characters in total. Send the chat history with every request; the final message must be a nonempty user question. Browser keys' system and developer messages are dropped; only server keys may add instructions. The model ID selects a project, not an upstream inference model; an unknown one returns 404 model_not_found. To choose how fast or deep the answer is, send a chat tier as docslurp.tier (fastest, balanced or smart); DocSlurp manages the model behind it. The endpoint uses the project's chat defaults.
Both JSON and streaming responses are supported. Set stream to true for server-sent content deltas, ending with [DONE]. With stream_options.include_usage, an extra chunk contains usage and an empty choices array. Usage reports actual answer-generation tokens; retrieval usage is recorded separately.
Set max_completion_tokens, or the max_tokens alias, up to 16,384; a key's chat output cap can lower it. DocSlurp runs its own project tools: tool_choice "none" asks for a direct answer, "auto" (the default) lets the search agent work, and client-defined tools are rejected. Images, audio, files, n above 1, and structured-output options are not supported. Unsupported requests return a descriptive error rather than silently changing behavior.
Answers include a Markdown source list. A DocSlurp-specific docslurp response extension also carries structured citations and availability. If indexing is still running, answers use only the documents indexed so far. The standard completion fields work with clients that ignore extensions. A docslurp request field (extra_body in OpenAI SDKs) takes the chat tier (docslurp.tier) and reasoning level (docslurp.reasoning), per-request retrieval overrides and speedBoost; see the search and chat API.
This implements project chat through Chat Completions and model discovery. Clients requiring the Responses API, tool execution, or other OpenAI endpoints need a different integration. For uploads and search, see the SDK quickstart.
Glossary
Open the explainer →