May Habib came to enterprise AI through a distinctly unusual path: Lebanese-born, Harvard Economics, then Lehman Brothers — where she worked through the 2008 collapse — then deploying billions in sovereign wealth fund capital for Abu Dhabi. When she and Syrian computer scientist Waseem AlShikh founded Qordoba in 2015 as an AI content localization tool, it wasn’t a direct leap to large language models. It was a decade of watching enterprise software be oversold and underdelivered.

That background shaped how they built Writer. The company’s thesis isn’t primarily technical. It’s organizational: enterprises that depend entirely on third-party model APIs are at structural risk. If OpenAI changes its pricing, deprecates a model, or makes a capability decision that conflicts with a customer’s use case, that customer has no recourse. Writer’s answer is to own the stack — proprietary models, proprietary retrieval, proprietary agent infrastructure — so that the enterprise relationship doesn’t hinge on a single upstream vendor.

Whether that argument holds up under competitive scrutiny is what this review examines.

The Origin: From Qordoba to Writer

Qordoba (incorporated 2015) was a content localization and style-consistency platform powered by AI. Habib and AlShikh raised approximately $23 million from Aspect Ventures and others and built a product that helped enterprises maintain brand voice at scale across languages. It was useful, moderately successful, and almost immediately eclipsed by what was coming.

The 2017 publication of “Attention Is All You Need” — the paper that introduced the Transformer architecture — changed what was possible for language models. Habib and AlShikh recognized that Qordoba was solving a narrow version of a problem that Transformers could address much more broadly. In August 2020, they pivoted. New company, new name, new product category: full-stack generative AI for the enterprise.

The Palmyra model family name is a direct tribute to the founders’ origins. Palmyra is an ancient Syrian city, a UNESCO World Heritage Site — and a deliberate signal that the models are built by people with roots outside the standard Silicon Valley pedigree.

Waseem AlShikh (CTO) won first place nationally in Syria’s computer science competitions in 2003 and 2004, then studied computer science and semantics at Beirut Arab University. He and Habib initially connected on Twitter around 2013 when she reached out about his work in statistical machine translation. A decade later, they had co-founded two companies together.

Funding and Business Scale

Writer’s capital history tells a story of methodical, milestone-driven growth — not the “raise as much as possible” pattern common in AI infrastructure.

Round Date Amount Key Investors
Seed 2020 ~$5M Aspect Ventures, Upfront Ventures
Series A 2021 ~$21M Insight Partners
Series B September 2023 $100M ICONIQ Growth (lead), Insight Partners, Balderton Capital, Accenture Ventures
Series C November 2024 $200M Premji Invest, Radical Ventures, ICONIQ Growth; plus Adobe Ventures, Citi Ventures, IBM Ventures, Salesforce Ventures, Workday Ventures

Total raised: approximately $326 million. The November 2024 Series C valued Writer at $1.9 billion — unicorn status, achieved in the same month.

The investor composition in the Series C is notable: Adobe, Salesforce, Workday, and IBM all invested. These are not passive financial bets — these are potential distribution partners and integration anchors for Writer’s enterprise platform. When Salesforce puts money in, it’s signaling that Writer is a plausible addition to the Salesforce ecosystem, not a competitor to be feared.

Revenue trajectory:

  • 2022: ~$2M ARR
  • 2023: ~$16M ARR
  • Late 2024: ~$47M ARR, growing approximately 194% year over year
  • Customers: 300+ enterprise accounts at Series C close

Writer has not disclosed more recent ARR figures. With a $1.9B valuation and ~$47M ARR, the implied ARR multiple is roughly 40x — aggressive, but defensible given the growth rate and the expanding agent platform revenue.

Employee count: Approximately 400 full-time employees, with offices in San Francisco (HQ), New York, London, and international expansion into Singapore, Dublin, Chicago, and Austin announced in early 2025.

The Palmyra LLM Family

Writer’s proprietary models run across several families with distinct purposes.

Palmyra X: The Frontier Models

The Palmyra X line is Writer’s competitive stake in the frontier model race.

Palmyra X4 (released October 2024) was notable for one specific reason: at release, it topped Berkeley’s Tool Calling Leaderboard, outperforming OpenAI, Anthropic, Meta, and Google on that benchmark. Tool-calling — the ability to reliably invoke external APIs and structured outputs — is critical for agentic enterprise use cases. Leaderboard rankings shift constantly, but landing at #1 at launch was a meaningful statement of technical intent.

Palmyra X5 (released April 28, 2025) is the current flagship. Its headline feature is a 1 million token context window — enough to process a substantial document corpus in a single prompt. Writer claims:

  • Full 1M-token prompt processed in approximately 22 seconds
  • Multi-turn function-calling latency of approximately 300 milliseconds
  • Pricing: $0.60/M input tokens, $6/M output tokens — Writer positions this as 3–4x cheaper than comparable GPT-4.1 deployments for long-context use cases
  • Available on Amazon Bedrock (as AWS’s exclusive cloud distribution partner at launch)

Benchmark claims for Palmyra X5 (Writer’s own reporting or third-party benchmarks where attributed): MRCR 8-needle long-context retrieval of 19.1% (slightly below GPT-4.1’s 20.25% on the same benchmark), BBH at 70.99%, MMLU_PRO at 65.02%, MATH_HARD at 71.57%, LongBench v2 at 53% average.

These numbers are published with appropriate caveats: benchmarks are snapshots, third-party validation is limited, and Writer naturally presents its model favorably. The 1M-token context window is the differentiated claim — most enterprise workflows won’t use it, but for document-intensive regulated industries (legal review, clinical trial analysis, financial compliance), it eliminates chunking and retrieval overhead that creates error surfaces.

Domain-Specific Models

Writer has published specialized models for regulated verticals:

Palmyra-Med (70B parameters) — fine-tuned for healthcare and biomedical language. Available on Hugging Face under an open license, used in clinical documentation and healthcare AI workflows.

Palmyra-Fin — fine-tuned for financial services language, supporting regulatory and compliance documentation workflows for customers like Franklin Templeton.

Open-Source Releases

Writer has open-sourced portions of its model portfolio:

  • Palmyra-Small (128M parameters) and Palmyra-Base (5B parameters) — original 2023 releases, available on Hugging Face
  • Palmyra-mini family (released 2025): three small models including Palmyra-mini 1.7B (base), Palmyra-mini-thinking-a 1.7B (reasoning), and Palmyra-mini-thinking-b 1.5B (math). Notably, Palmyra-mini is the first Writer model to run on-device on iPhone via llama.cpp on iOS 18.

Open-sourcing small models while keeping frontier models proprietary is standard competitive practice — it builds developer goodwill and ecosystem presence without giving away the core product.

The Knowledge Graph: Beyond Vector RAG

Writer’s most technically distinctive claim is its Knowledge Graph retrieval system, which it positions as superior to conventional vector-similarity RAG.

Standard RAG works by embedding documents as vectors, then retrieving the chunks most similar (by cosine distance) to an incoming query. This approach has a known limitation: vector similarity is not logical reasoning. Two documents may be semantically distant from a query but jointly contain the answer. Multi-hop reasoning — “based on policy X from 2022, how does that interact with the 2024 revision of process Y to affect customer Z?” — is where vector RAG tends to fail.

Writer’s Knowledge Graph uses a specialized LLM to parse enterprise data and build a semantic graph structure, where nodes represent data points and edges represent the relationships between them. Queries navigate this graph following relational paths, not distance metrics. This enables multi-hop reasoning natively.

Performance claim: In a benchmarking study, Writer’s Knowledge Graph achieved top performance on RobustQA, outperforming seven popular vector-RAG approaches. This is a vendor-conducted study — readers should treat it as directionally indicative, not independently validated.

Practical implications for enterprise customers:

  • Customer service agents can follow policy relationships across document versions
  • Financial compliance tools can trace regulatory lineage without manual cross-referencing
  • Sales enablement can answer relationship-dependent questions across large product and pricing catalogs

Whether the Knowledge Graph’s practical advantage over well-tuned vector RAG is significant in production remains an open question. Graph-based retrieval also has costs: building and maintaining the graph requires ongoing inference spend, and schema changes in enterprise data can require substantial re-indexing. Writer doesn’t publish detailed total-cost-of-ownership comparisons.

AI HQ: The Agentic Platform

Writer launched AI HQ on April 10, 2025, positioning it as the primary interface for enterprise AI agent deployment. It became generally available for all customers on June 25, 2025.

AI HQ has three main components:

Agent Builder — A low-code drag-and-drop interface for constructing agents from modular blocks. Both IT teams and business users (HR, finance, marketing) can build without writing code. Blocks include tool-calling, Knowledge Graph retrieval, LLM inference, conditional logic, and output formatting. Pre-built integrations connect to Adobe, Salesforce, Workday, Microsoft, and Atlassian ecosystems.

Agent Library — 100 ready-to-use pre-built agents covering common workflows across finance, healthcare, retail, and technology. These function as starting points that customers can clone and customize rather than building from scratch.

Observability & Governance — Real-time visibility into deployed agent activity. IT administrators can monitor all agents, detect anomalies, and intervene on issues from a central dashboard. This is a direct response to enterprise IT’s core concern with autonomous agents: auditability and control.

Beta customers at launch: Uber, Salesforce, Franklin Templeton, and Commvault — a strong enterprise validation set. Uber in particular is notable: their AI HQ deployment handles content freshness across a 40,000-person global support team, automatically updating CMS content based on operational changes.

The framing for AI HQ — “build, activate, and supervise AI agents” — is specifically designed to address procurement objections. Most enterprise AI deployments stall not on technical capability but on governance: who approved this agent to run? What data did it access? Who is accountable when it produces an error? AI HQ’s observability layer is built to answer those questions for auditors, compliance teams, and IT security.

Enterprise Security and Compliance

Writer’s compliance posture is competitive with the most credentialed AI vendors in the market:

  • SOC 2 Type II (annual audits covering Security, Availability, and Confidentiality)
  • ISO/IEC 27001 (information security management system)
  • ISO/IEC 27701 (privacy information management)
  • ISO/IEC 42001 (AI management system — the recently introduced AI-specific standard that only a handful of providers have achieved)
  • HIPAA/HITECH compliant

The no-training-on-customer-data commitment is stated explicitly in Writer’s terms of service and is cited by enterprise customers as a critical procurement requirement — particularly in regulated industries where data minimization is not optional.

Deployment architecture: Writer’s primary model is SaaS/cloud. Enterprise customers with strict data residency requirements should verify current VPC/private cloud options directly with Writer’s sales team, as public documentation on exact deployment modes is not fully detailed for all configurations.

MCP Integration

Writer has committed meaningfully to the Model Context Protocol ecosystem:

Official MCP Server (writer-sdk-mcp, available on npm): Enables any MCP-compatible client (Claude Desktop, Cursor, Windsurf, etc.) to call Writer APIs through the MCP protocol. Exposes tools for file upload and analysis, content generation via Palmyra models, Knowledge Graph queries, and image analysis. Requires a WRITER_API_KEY environment variable. Writers’s MCP server is a genuine developer-facing offering, not an afterthought.

Enterprise MCP Gateway (in development): Writer is building a gateway layer that sits between enterprise agents and external MCP servers, adding security controls, governance, audit logging, and access management that the base MCP protocol doesn’t provide. The engineering rationale is sound: the MCP protocol’s design deliberately doesn’t address enterprise requirements like access control and audit trails; a gateway fills that gap.

Writer’s engineering team has also published on technical MCP deployment patterns — “avoiding context rot” in agent systems, auto-generating MCP tools from OpenAPI specs and Postman collections — suggesting genuine technical investment rather than marketing-driven positioning.

Key Customers and Use Cases

Uber — Writer agents keep CMS content current across a 40,000-person global support team. When operational details change, agents automatically update the relevant documentation without requiring manual editorial cycles.

Franklin Templeton — Financial advisors generate market commentary and portfolio analysis grounded in internal Snowflake data and SEC filings. Compliance controls verify that outputs meet regulatory standards before distribution.

L’Oreal — Digital marketing content generation at global scale, with Writer maintaining brand voice consistency across product lines and markets.

Spotify — Content workflow automation.

Vanguard — Enterprise deployments, with Vanguard having invested in Writer’s Series B.

Accenture, Intuit, Mars, Marriott, Commvault — all confirmed customers across consulting, financial software, consumer goods, hospitality, and data management.

The customer list is weighted toward large enterprises in regulated or high-volume-content industries — precisely the segment where the full-stack argument is strongest. A law firm managing thousands of client documents, a pharmaceutical company running clinical document review, or a financial services firm processing regulatory filings can justify the cost and complexity of Writer’s platform better than a startup that just needs a model API.

Competitive Positioning

Writer vs. OpenAI Enterprise

OpenAI Enterprise is primarily a frontier model API with a managed interface (ChatGPT Enterprise). Writer provides a richer application layer, purpose-built enterprise governance, domain-specific vertical models, and a Knowledge Graph. Writer’s “no training on your data” commitment is also more prominently featured and contractually explicit than OpenAI’s equivalent statements, which matters in regulated industries.

The counterargument: GPT-4o and GPT-4.1 have broader benchmark coverage and more third-party validation than Palmyra. Enterprises that need a model with extensive public track records lean toward OpenAI.

Writer vs. Anthropic Claude for Enterprise

Similar dynamic. Anthropic is a model company; its enterprise offering is Claude.ai for Teams/Enterprise plus the API. Writer has a substantially deeper pre-built application layer and stronger vertical compliance certifications. Anthropic’s advantage is Claude’s documented reasoning capability and the trust generated by its safety research program.

Writer vs. Cohere

This is the most direct structural comparison. Both are enterprise-first, own-model AI companies with strong data privacy commitments and serious compliance certifications. Both are building agentic platforms. The key differences:

  • Cohere has raised more (~$1.54B before the Aleph Alpha merger), has historically stronger documentation of flexible deployment including on-premises, and has a differentiated retrieval story (Embed, Rerank) that’s well-established in the vector RAG ecosystem. Cohere’s merger with Aleph Alpha creates a transatlantic sovereign AI story.
  • Writer has a richer application/agent layer (AI HQ), stronger compliance pedigree (ISO 42001, HIPAA), a Knowledge Graph approach to retrieval, and 1M-token context on Palmyra X5. Writer also has a more developed sales integration through its Series C investor relationships (Salesforce, Workday, Adobe).

Enterprises evaluating both should test their specific use case. Cohere wins on flexibility and retrieval maturity; Writer wins on application completeness and governance tooling.

Writer vs. AI Wrappers (Jasper, Copy.ai)

Writer explicitly rejects comparison to AI-wrapper companies that build on closed third-party models. The business risk is real: if OpenAI raises prices 4x or discontinues a model, Jasper has no recourse. Writer’s proprietary model ownership is structural protection against that scenario. This argument resonates most with procurement teams that have experienced vendor lock-in elsewhere in their software stack.

What Doesn’t Work Yet

Benchmark opacity. Many of Writer’s performance claims come from Writer’s own benchmarking studies or Writer’s own announcements. The Palmyra X4 Tool Calling Leaderboard result is independently sourced, but it’s from 2024 and rankings have shifted since. More third-party, independent evaluation of Palmyra X5 would strengthen confidence.

Knowledge Graph total cost. Writer doesn’t publish clear documentation of the cost and operational overhead of maintaining a Knowledge Graph versus well-tuned vector RAG. For many enterprise use cases, the Knowledge Graph advantage may not justify the additional complexity, but Writer’s marketing doesn’t help customers evaluate that tradeoff.

Deployment flexibility uncertainty. While Writer’s SaaS platform is well-documented, enterprises with strict data sovereignty requirements (government, defense, certain healthcare scenarios) need clearer VPC/private deployment options. Cohere has historically been more explicit on this dimension.

Developer ecosystem maturity. Writer’s developer portal and community are growing, but are not yet at the breadth of Anthropic’s or OpenAI’s developer ecosystems. The Palmyra-mini open-source release is a step toward building developer goodwill, but the gap remains.

Bottom Line

Writer’s full-stack bet is coherent. Owning the model, the retrieval layer, the agent infrastructure, and the compliance certifications eliminates the structural dependency on upstream AI vendors that threatens pure “wrapper” products. For large enterprises in regulated industries — financial services, healthcare, legal, large-scale consumer operations — that argument makes practical sense.

The platform has real enterprise validation. Uber deploying Writer across 40,000 support team members, Franklin Templeton building regulatory-grade financial commentary tools, Salesforce and Workday both investing in the Series C — these aren’t vanity metrics. They’re evidence of enterprise trust at scale.

The current limitations are real but tractable: benchmark transparency could improve, deployment flexibility documentation could be clearer, and the developer ecosystem needs continued investment. None of these are structural problems.

Rating: 4/5. A credible full-stack enterprise AI platform with genuinely differentiated architecture, serious compliance credentials, and enterprise customers that validate the value proposition. Deductions for limited third-party benchmark validation of Palmyra and incomplete public documentation of deployment options.

This review is based on publicly available information as of May 2026. ChatForest conducts desk research only — we have not accessed Writer’s platform or API for this review. Enterprise AI capabilities evolve rapidly; verify current specifications with Writer directly before procurement decisions.