Writer
Listed at https://writer.com
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | Palmyra X5: $0.60/M, Palmyra X4: $2.50/M | per million tokens |
| Output | Palmyra X5: $6.00/M, Palmyra X4: $10.00/M | per million tokens |
🤖 Supported Models (2)
✨ Pros
- ✓Palmyra X5 flagship, 1M context window
- ✓Enterprise AI platform: Knowledge Graph RAG, Agent Builder, Guardrails
- ✓OpenAI-compatible API format
- ✓Streaming, structured output, function/tool calling
- ✓Vision support (images/video)
- ✓SOC 2 certified, enterprise security compliance
⚠️ Cons
- ×Limited model selection (only 2 active models)
- ×China access stability unverified
- ×No free API credits (14-day trial only)
- ×Palmyra X4 less competitive vs alternatives
- ×No public affiliate program
🎯 Best For
Enterprise AI applications; Knowledge Graph RAG and agent workflows; compliance-heavy use cases
💰 Pricing & Plans
| Plan / Model | Price | Included Limits | Best For |
|---|---|---|---|
| Palmyra X6 (`palmyra-x6`) | $2.00 input / $8.00 output per 1M tokens | 1M token context, 8,192 max output; default model for WRITER Agent; about $0.12 per finished task at 26 s median latency | Long-horizon agentic work: grounded research, multi-step playbooks, connector actions |
| Palmyra X5 (`palmyra-x5`) | $0.60 input / $6.00 output per 1M tokens | 1M token context, 8,192 max output; deprecated 2026-12-14, migrate to X6 | Cheapest Palmyra tier for high-volume extraction, classification and summarisation |
| Palmyra X5 inside WRITER Agent | $5.00 input / $12.00 output per 1M tokens | Same model, agent-surface surcharge over the raw API rate | Agent sessions only - the API / Agent Builder / no-code rate stays at $0.60 / $6.00 |
| Palmyra X4 (`palmyra-x4`) | $2.50 input / $10.00 output per 1M tokens | 128k context, 4,096 max output; deprecated 2026-12-14 | Legacy Bedrock ID `writer.palmyra-x4-v1:0`; migrate to X6 (1M context) |
| Starter plan | 14-day free trial, no credit card; then monthly or annual per seat with fixed credit limits (seat price not published online) | Up to 5 users; up to 5 Playbooks; 1 team personality profile; basic connectors; 50 GB storage in a single Knowledge Graph; 20% nonprofit / education discount | Fast-moving teams kickstarting AI exploration |
| Enterprise plan | Contact sales | Unlimited users, playbooks, routines and cross-team workflows; unrestricted Knowledge Graph connectors; SAML SSO, SCIM, granular RBAC, audit logs; guardrails; dedicated private-cloud deployment; HIPAA / SOC 2 Type II reports on request | Large organisations scaling agentic workflows with compliance requirements |
| Credits (consumption model) | 1 credit = 1 USD, held as an org-level dollar balance; Lite seat = $30 per user per month; Pro seats draw from the shared pool | Credits power WRITER Agent, prebuilt agents, playbooks, API and AI Studio; platform and Pro credits expire at contract end, Lite credits reset monthly and do not roll over; Consumption view reports usage from July 2026 | Metering layer that sits under every WRITER surface |
| Knowledge Graph add-ons | Hosting $0.085 per GB of storage per day; data extraction $0.00015 per page; file parsing (OCR) $0.055 per page; Web Access $0.12 per page | Graph-based RAG with inline citations; data connectors for Knowledge Graph are Enterprise-only | Grounded, citable answers over your own documents and websites |
| WRITER-managed third-party models | Per 1M tokens: Claude Opus $5.00 / $25.00, Claude Sonnet 5 $2.00 / $10.00, DeepSeek V4 Flash $0.09 / $0.18, GPT-OSS 120B $0.04 / $0.18, GLM 5.1 $0.98 / $3.08, Veo 3.1 $0.40 per second | WRITER-metered catalog available in WRITER Agent only; rates last updated 2026-09-17 | One endpoint with many models, billed through a single WRITER credit pool |
| BYOM (bring your own model) | Your provider rate card plus a 10% WRITER pass-through fee, paid from credits | Connect AWS Bedrock or OpenRouter with your own credentials; Enterprise plans only | Keep your own vendor billing while keeping Writer's orchestration layer |
| Tools API - PDF parser | $0.055 per page | Pre-processing utility for understanding text before it reaches a model | Document pipelines that need parsed text rather than raw PDF bytes |
| Palmyra Vision (deprecated) | Image $0.005 per image, video $0.005 per second, text $7.50 input / $22.50 output per 1M | Deprecated 2026-07-13; use chat-with-images on Palmyra X5 instead | Legacy multimodal workloads only |
🔧 API & Developer Experience
- •REST surface: POST https://api.writer.com/v1/chat for chat completions plus /v1/completions for single-shot generation; responses add Writer-specific fields (graph_data, llm_data, image_data) on top of the OpenAI-style shape; streaming via Server-Sent Events.
- •Official SDKs: Python `pip install writer-sdk` (`from writerai import Writer`), Node `npm install writer-sdk` (`import { Writer } from 'writer-sdk'`); the client auto-reads the WRITER_API_KEY environment variable.
- •Structured outputs and tool calling: json_schema response_format on /v1/chat, plus SDK parse helpers for Pydantic and Zod; custom function calling, plus built-in tools for web search, model delegation and Knowledge Graph queries.
- •Knowledge Graph, files and tools: Graph RAG API with inline citations, KG query configuration, web URL connector; File API for upload / list / delete; PDF parser tool billed at $0.055 per page.
- •Agents, playbooks and Agent Builder v2: Applications API turns deployed no-code agents into microservices, with an async job API for batch runs; Agent Builder v2 (open beta) deploys your own dockerised FastAPI, Next.js, LangChain, LangGraph, CrewAI, Letta, LlamaIndex, Strands, Agno or OpenAI Swarm app with dev / prod environments and GitHub sync.
- •MCP server and enterprise connectors: Hosted WRITER MCP server at https://api.writer.com/v1/mcp works with Cursor, Claude and Grok Bot; custom connectors can be built from OpenAPI specs or MCP servers, with private endpoints over AWS PrivateLink and GCP Private Service Connect.
- •Rate limits, keys and governance: Default 400 RPM and 25,000 TPM, custom limits by sales; API keys are organisation or user scoped with per-capability permissions and expiry, and legacy keys stop working 2026-12-01; Control Center exposes event logs, a Logs Export API, an Audit Events Export API and a per-agent consumption view.
🧠 Knowledge Graph RAG & Governed Agents
Writer is not a raw model vendor - it is an enterprise agent platform whose flagship Palmyra models were co-developed with its orchestration layer. Its Knowledge Graph builds a semantic network of entities, concepts and relationships from your documents, websites and connectors, so answers arrive with inline citations instead of keyword matches, and grounded prompts behave like retrieval rather than guesswork. On top of that sits a full agent layer: playbooks, sub-agent delegation, MCP tool use across 25+ connectors (Salesforce, HubSpot, Slack, Snowflake, Notion, Confluence), guardrails that screen PII and prompt injections before and after the model call, and audit logs for compliance review. Palmyra X6 is priced for finished tasks rather than raw tokens alone: roughly $0.12 per completed task at a median 26-second latency. Grounding, orchestration and governance on one surface is what separates Writer from a plain per-token API.
🌐 China Access & Latency
Writer publishes no mainland China region, no ICP filing and no Chinese data-residency guarantee; the platform runs on AWS with dedicated private-cloud deployments, and full data residency is an Enterprise-contract item. From our Alibaba Cloud Malaysia probe, api.writer.com sits behind Cloudflare anycast (104.20.31.216 / 172.66.167.93, cf-ray edge SIN) and answered every unauthenticated request with a hard 403 WAF block, because the traffic originated from a datacenter IP. Expect the same friction from China-hosted cloud egress, CI runners and cheap VPS hosts - the block is IP-reputation based, not a China-specific ban. Documentation on dev.writer.com (Vercel) returned 200 with 83-108 ms TTFB, so the product and docs pages do load. The practical route for mainland teams is to call Palmyra through Amazon Bedrock model IDs writer.palmyra-x5-v1:0 and writer.palmyra-x4-v1:0, or to bring your own OpenRouter and Bedrock accounts via BYOM, then test egress early rather than after integration.