citellm.com

What It Is

CiteLLM positions itself as an “LLM verification layer”: users submit a PDF and an extraction schema, and the system returns structured JSON with a page number, PDF coordinate box, source text snippet, and confidence score for each field. The problem it addresses is not simply extraction speed, but traceability — “where did this value come from?” This makes it especially relevant for human review, audit, and appeal workflows in regulated industries.

Core Capabilities

Its core capabilities include field-level citations, click-to-highlight navigation, confidence scoring, schema-driven extraction, historical document/extraction management, and embeddable React/JS verification components. The API is relatively straightforward: POST /v1/extract, with PDFs passed in via Base64, URL, or document_id. Typical use cases include loan income verification, contract clause extraction, insurance claims, M&A due diligence, SEC filing analysis, and invoice processing. The documentation also provides recommended confidence ranges, indicating that the product is designed to encourage human-in-the-loop workflows by default rather than fully automated decision-making.

Pricing and Deployment

Public pricing lists Starter at $99/month including 1,000 pages, and Growth at $499/month including 8,000 pages with additional pages at $0.08 each; Enterprise pricing is custom. Deployment options include a Cloud API, Docker self-hosting, on-prem deployment, and support for air-gapped environments, which is useful for sensitive documents. However, free quota, trial policy, enterprise SLA, and payment methods are not disclosed and require requesting access.

Pros and Cons

The main strengths are its clear evidence-chain design and coordinate-level citations, which can significantly reduce the cost of reviewing long PDFs. Its API, SDKs, and self-hosting path are also fairly complete, making it a good fit for high-compliance scenarios such as finance, legal, and insurance. The drawbacks are that it does not specify the underlying model, OCR capabilities, accuracy benchmarks, or performance on complex layouts. For Chinese support, only a language auto parameter is visible, with no explicit quality commitment for Chinese. The product currently appears to focus mainly on PDFs, and support for other document formats is unclear.

Who It’s For and Access from China

CiteLLM is suitable for B2B teams that already have document AI workflows but struggle to put them into production due to auditability and trust concerns. This is especially true for lending platforms, law firms/legal teams, insurance operations, finance AP teams, and investment research/compliance teams. The source material does not provide details on access from China, so testing is required. If network, compliance, or payment limitations apply, alternatives such as Azure AI Document Intelligence, Google Document AI, AWS Textract, or a localized OCR + LLM solution may be worth evaluating.