Inkeep

What it is

Inkeep is an agent platform for customer experience and support teams that you build in either of two ways: a drag-and-drop visual builder for support and content people, or a TypeScript SDK for engineers, with two-way sync so an agent edited on the canvas shows up as a code change and vice versa. It grew out of a docs-grounded “ask AI” widget for developer-tool companies and now ships three agent types: customer-facing assistants on your docs and help centre, internal coworker agents that answer support reps inside Slack or Zendesk with context pulled from CRM and billing, and background automation agents that fire on events to update knowledge bases, file tickets, or write CRM records. Agents call tools over MCP, talk to each other over an A2A API, and emit OpenTelemetry traces you can inspect.

The framework is on GitHub as inkeep/agents under the Elastic License 2.0 plus Inkeep’s own supplemental terms, and self-hosts on Vercel or Docker. The company is San Francisco-based, part of Y Combinator’s W23 batch, founded by Nick Gomez (CEO) and Robert Tran (CTO), and raised a $13M seed led by Khosla Ventures, GreatPoint Ventures, and Y Combinator in September 2025. Named production customers include Anthropic, Midjourney, Clay, PostHog, Pinecone, Postman, and Clerk.

Why it shows up in Customer Success stacks

• You own and version the agent. Closed support agents (Intercom Fin, Sierra, Decagon) keep the agent definition inside the vendor. With Inkeep the agent is a TypeScript project in your repo: it goes through code review, diffs cleanly when a support lead changes a prompt on the canvas, and survives a vendor exit because the framework source is public.
• Built for technical products. Citation-first answers grounded in docs, GitHub, and community threads are the origin of the company, which is why the customer list reads like a developer-tools roster. If your tickets are “why does this API call return 403,” Inkeep is closer to the problem than a helpdesk-native bot trained on macros.
• One platform for the whole support loop. The same builder covers the public docs assistant, the rep-facing copilot in the Zendesk sidebar (draft a grounded reply, apply it to the ticket in one click), and the agent in Slack channels, so you maintain one knowledge setup instead of three vendors.

Pricing reality

Inkeep publishes no list price. Two paths:

• Open source, $0 licence. Visual builder, SDK, MCP tools, observability, and the UI component library, self-hosted. You pay for hosting, a database, and LLM tokens from the provider you choose, plus the engineer who runs it. Community support only.
• Enterprise, quoted. Adds managed hosting, the managed RAG search layer over your docs and help centre, the Slack and support-platform integrations, PII removal and data controls, SSO and user management, and a dedicated forward-deployed engineer. Inkeep says it quotes against your expected usage; there is no free cloud tier, only a 30-day demo sandbox.

What that means in practice: the older self-serve docs-assistant plans (listed at $150–$500 a month on third-party directories) are gone, and the forward-deployed-engineer model is an annual enterprise contract, not a credit-card purchase. The comparable managed docs assistant, kapa.ai, starts around $250 a month and runs past $500 a month on annual terms. Budget Inkeep Enterprise above that band, and get the quote broken out by conversation volume before you compare it with per-resolution vendors.

Best for

Support and developer-experience leads at developer-tool and API companies, roughly 50 to 1,000 employees, with an engineer who can own a TypeScript project, who want one agent layer across docs Q&A, rep assist in Zendesk or Slack, and knowledge-base upkeep, and who want that agent versioned in their own repo.

Not for a support team with no engineering time and a consumer or ecommerce ticket mix (order status, refunds, returns): a helpdesk-native agent priced per resolution gets you live faster and bills only on success.

Watch-outs

• “Open source” is source-available. ELv2 bars offering the software to others as a hosted or managed service and circumventing licence keys, and Inkeep layers supplemental terms on top. Guard: have counsel read the licence plus supplemental terms before you build a customer-facing product on it, and pin the version you ship.
• Self-host moves the ops burden to you. Uptime, upgrades, secrets for MCP credentials, and LLM spend become your team’s job, and the managed RAG search and support-platform integrations are Enterprise features. Guard: price the self-host path as hosting plus tokens plus the engineering hours to run it, and compare that total, not $0, with the Enterprise quote.
• Opaque pricing makes comparison hard. Without a published rate you can’t do per-resolution math against Fin or Zendesk. Guard: ask for the quote expressed per conversation at your current monthly volume, with the overage rate and renewal cap in writing.
• Seed-stage vendor. $13M raised is small next to Sierra or Decagon. Guard: request the current SOC 2 report from the Inkeep Trust Center, and keep your docs and help centre as the system of record so the agent layer is replaceable.

Alternatives

Intercom Fin is the market-share leader in AI support agents at about $0.99 per resolution; pick it when you want a closed agent live in days and are happy to let the vendor own the definition. Zendesk’s own AI agents are the other incumbent; pick them when you already pay for Suite and want one vendor. Sierra is the fastest-growing entrant; pick it (or Decagon) for high-volume, custom-priced enterprise deployments with voice and deep backend actions. If the use case is only docs Q&A with no agent building, kapa.ai is the narrower managed option. If you want general-purpose automation rather than a support-shaped agent platform, n8n or Relevance AI cover more workflow types. Pick Inkeep when your product is technical, your answers live in docs, and you want the agent in code you control.