Retab

Market position

Retab presents itself as infrastructure or middleware between unstructured enterprise documents and AI agents, targeting vertical AI startups and internal innovation teams [6][7]. Coverage frames it as a developer-first entrant addressing the gap between document AI demos and production reliability [3][6].

Company and press materials emphasize the orchestration layer rather than the model: founders concluded from building internal logistics automation tools that the durable value lay in the orchestration layer around models [6][7]. Claimed differentiators are model-agnostic routing (described as making processes up to 100x cheaper than other solutions), self-optimizing schemas, k-LLM consensus for uncertainty quantification, and end-to-end handling of the lifecycle from dataset labeling and evaluation to prompt engineering and model selection [6][7][3]. The website also claims "Retab split beats Google" [0][1]. These performance claims originate from the company and are not independently verified in the sources.

Technology

An LLM-first document pipeline that replaces legacy OCR/ETL stages with schema-driven extraction APIs [4]. Components cited include layout-, table- and schema-aware parsing and extraction; self-optimizing schemas refined by an agent against customer documents; model-agnostic benchmarking and routing across providers such as OpenAI, Google and Anthropic; guided step-by-step reasoning; and a k-LLM consensus mechanism producing per-field likelihood scores to quantify uncertainty [3][6][7][0]. The platform adds citations/source grounding, confidence scores, evaluations and monitoring, schema versioning, run inspection and human review queues, and exposes workflow context to agents through MCP tools [0][1][2].

Go-to-market

Retab uses a developer-led motion: an open-source GitHub repository, public documentation, a CLI installable via a shell script, SDKs for Python, Node, Go and Java, an MCP integration for AI agents, an API playground and a Discord community [2][4]. The website offers both self-serve signup and a demo request [0][1]. The company launched publicly out of stealth on 30 July 2025 with a press release and coverage in trade and startup media [6][7][3], and said it was launching integrations with automation platforms including n8n, Zapier and Dify [6][7].

Developers building production applications, plus operations, financial analysis and AI teams [0][1][6]. Named or described customers and sectors include logistics (a major trucking company), financial services (extracting quantitative metrics and qualitative risk factors from 200-page quarterly reports), healthcare, claims processing, medical records, identity verification and onboarding [6][7]. Logos displayed on the website include Harvard University, Sinari, Carmoola and Maersk [0][1]. At launch the company said the platform was used by dozens of companies [6].

Geography

Sources conflict: EU-Startups describes Retab as Paris-based [7], while the launch press release describes it as a San Francisco-based startup and is datelined San Francisco [6]. A third-party database lists the HQ as Paris, France [5].

History

Press coverage of the launch states the company was founded in 2023 [7][5], which conflicts with the internally held founding year of 2024. Retab emerged from stealth on 30 July 2025, simultaneously announcing a $3.5 million (€3 million) pre-seed round and the general launch of its platform [6][7][3]. A third-party aggregator instead lists two rounds totaling $7.0M ($4M in May 2024 and $3M in July 2025) with a different investor set, which is inconsistent with the primary press coverage [5].

Risks & controversies

Sources contain conflicting company facts: founding year (2023 per press and aggregator [7][5] versus 2024 internally), headquarters (San Francisco in the launch release [6] versus Paris [7][5]), and funding history (a $3.5M pre-seed from VentureFriends, Kima Ventures and K5 Global [3][6][7] versus an aggregator's $7.0M across two rounds with a largely different investor list, and an employee band of 201-500 that conflicts with the reported ten employees [5][6]). The launch article on Yahoo Finance is labeled a paid press release [6]. Performance claims such as 100x cost reduction and "Retab split beats Google" are vendor-stated and unverified [6][0]. The platform depends on third-party model providers [3][6].