Parallel AI

Parallel AI is a San Francisco-based artificial intelligence company focused on AI research tooling, web search capabilities for AI agents, and developer infrastructure. Jan Wilmake worked at Parallel AI as a Developer Relations (DevRel) contractor from approximately August to November 2025.^[1]

Products

• Parallel Search MCP – A Model Context Protocol (MCP) server providing web search and web fetch capabilities for AI agents
• Context tooling – Systems for managing and retrieving context within LLM workflows
• Cloudflare integrations – Edge-compute deployments leveraging Cloudflare Workers

Jan Wilmake's Role

Contract negotiations were finalised in late July 2025 after Wilmake needed several weeks to review the initial terms.^[2] Wilmake joined at approximately 32 hours per week and structured his working week around Parallel's schedule: Monday–Thursday focused on Parallel deliverables, with a longer weekend for personal product work.^[2] His work focused on:

• Developing and documenting MCP server integrations
• Building Cloudflare-based infrastructure components
• Creating developer-facing examples and integrations
• Contributing to context tooling and API documentation

Wilmake also pitched Cloudflare engineer Kristian Freeman on building a standalone MCP client — framed as "a T3 Chat but for MCP" — as a showcase for Cloudflare's developer platform, though this proposal was not formalised.^[2]

During the same period (July 2025), the open-source email project mail0 approached Wilmake about a separate Cloudflare Durable Objects contract, creating overlapping opportunities.^[2]

The engagement ran August–November 2025, ending when Wilmake experienced a health episode in Vietnam (see Mental Health History).^[1]

Relocation Plans

During the Parallel AI period, Wilmake and his fiancée Bruna began actively planning a relocation to the San Francisco Bay Area, specifically targeting Woodside or other hillside communities near Palo Alto for proximity to nature and the Bay Area tech network.^[2]

Model Context Protocol (MCP)

A significant portion of Wilmake's work at Parallel AI involved the Model Context Protocol (MCP), an emerging standard for connecting AI models to external data sources and tools. This work extended technical expertise developed through uithub and ActionSchema.

Parallel AI operates as a listed MCP server provider on claude.ai, providing web search and web fetch capabilities to Claude users.

Employment Context

The Parallel AI role followed a period of declined offers in mid-2025:^{[1] [3]}

• May 2025: A trial week at Iterate.com (~$3,600 earned), offer declined
• May 2025: Multiple other offers declined — Skyfire, ComposioDev, StackOne, SocialDatabase
• May 2025: Investor Ben Tossell expressed interest in backing Wilmake's work
• May 2025: Upstash began sponsoring uithub at $100/month
• Ongoing: Cloudflare DevRel partnership negotiations, stalled
• August 2025: Parallel AI DevRel contract begins

Total 2025 earnings were approximately €61,000, with Parallel AI contributing approximately $67,200 USD across several contract periods.^[4]

2026 Strategy and DevRel Niche

In January 2026, Jan published a retrospective and forward-looking post on his role at Parallel. Key findings from 2025: 2.5M X impressions, +2,600 GitHub stars across new repos, and a significant body of work (cookbook recipes, n8n integration, OAuth provider, Task MCP, LLMTEXT). The main challenge identified: sustaining X traffic and converting it into meaningful Parallel product adoption.

His 2026 niche within DevRel: the intersection of code (Cloudflare + Parallel) and indie hackers / playful mindset, specifically targeting developers who build in public, SaaS builders, and web scraping communities. He contrasted this with the no-code / enterprise quadrant, which he left to others.^[5]

His 2026 plan had three components:

1. Free tools and DevTools — continuing LLMTEXT and cookbook-style tools that generate genuine developer value and funnel users toward Parallel's paid APIs (product-led growth)
2. Wilmake Stack — a skills library making it easier to build apps with Parallel primitives
3. Build SaaS for the agentic internet — building a profitable independent business using MCP and agent standards, then selling it; ideas included agent-friendly web (markdown browser), socialdatapod, OSS jobs finder, and a context marketplace

AI Inference DX Design

Concurrently with the Parallel engagement, Jan was building contextarea — a developer-experience layer on top of AI inference. A 2025 draft post describes the challenge: around every model's core token-in/token-out function, there are many desirable features (response caching, billing, identity, URL context, MCP tool use, code mode, cronjobs, async execution), and composing them into a coherent developer experience is "more of an artform than simply following rules in a book."^[6]

This work sits at the intersection of Parallel's infrastructure goals and Jan's own product interests in MCP and agent tooling.

October 2025: OpenAI Apps SDK and Parallel MCP UI

In early October 2025, OpenAI released the Apps SDK — an MCP-based standard for building MCPs with interactive UIs. Jan built a UI for the Parallel MCP using this SDK, working through initial bugs before getting it functional.^[7]

Jan also posted about the need for a standard for how MCP Clients can expose uploaded files to MCPs — noting that existing clients (Claude, ChatGPT, IDE-based) all handle file uploads differently and inefficiently, inserting file content token-by-token into the context window.^[8]

October 2025: Parallel Task API and Entity Resolution

In mid-October 2025, Jan worked on using the Parallel Task API for entity resolution — finding all social media profiles belonging to a person given limited identifying information. He designed a shape for a person entity resolution API and tested it with Parallel's core processor.^[9]

On December 9, 2025, Jan shipped an entity resolution feature: "find all social media profiles belonging to any person, given enough information — name, email, username, or any profile URL — returns matching profiles across platforms for sales, recruitment, and data quality use cases."^[10]

November 2025: Search API Launch

On November 6, 2025, Parallel AI launched the Parallel Search API — described as "the most accurate web search for AI agents, built using our proprietary web index and retrieval infrastructure." Jan announced the launch on X, writing: "today we're launching the new Search API at @p0; the highest accuracy web search API engineered for AI... by engineering every layer of the stack from first principles our team has achieved state-of-the-art results."^[11]

The benchmark results from CEO Paraga_Agarwal showed Parallel outperforming products built over years by major companies. Jonas Templestein publicly switched from ChatGPT to Claude to Parallel for deep research the same week, calling each step a major improvement.^[12]

January 2026: Parallel Task MCP Update

On January 15, 2026, Jan announced that the Parallel Task MCP now also supports "fast" processors — expanding the available processing tiers.^[13]

January 2026: Parallel on Google Vertex AI

On January 14, 2026, Parallel announced that grounding with Parallel web search was available in preview on Google Vertex AI.^[14]

January 2026: Monitor API Structured Outputs

On January 22, 2026, Parallel announced structured outputs for the Parallel Monitor API — always-on web searches with configurable schemas for structured responses.^[15]

February 2026: Parallel as DevRel Showcase

Jan used Parallel's Extract API as the backend for Markdown_Browser (launched March 2, 2026), where sites that only serve HTML (without native markdown) require a Parallel API key to render. This positioned Markdown Browser as a direct showcase for Parallel's Extract API capability.^[16]

March 2026: Markdown Browser and Defuddle

Jan added pluggable extract engines to Markdown Browser in March 2026, with Defuddle (by @kepano) becoming the default. Parallel's API remained available as an option for HTML-only sites. Jan tested edge cases — e.g., parallel.ai/blog only showed the top blog post with Defuddle but worked better with Parallel's API.^[17]

April 2026: Free MCP Search Launch

On April 24, 2026, Parallel AI launched free web search for AI agents via MCP — requiring no account, no API keys, and zero cost. The announcement tweet from @p0 stated: "The best web search for agents is now free. Upgrade to Parallel's web search tools in any MCP-supported tool or agent, for free, in under 60 seconds." Utkarsh Srivastava (@utkarsh) followed up: "The same quality that enterprises rely on, now available to individual developers for free."^[18]

Jan read the announcement on April 24 and revisited it on April 26. Paraga_Agarwal also posted about major upgrades across the board and global coverage for the Search and Extract APIs, which Jan liked on X.^[19]

This launch is strategically significant for Jan's product portfolio: Self-Driving_Repos uses Parallel's web search as a tool for its agents. With the search layer now free, the marginal cost of each agent run that includes web search drops to zero — improving the economics of the platform. The free MCP search also lowers the barrier for any MCP-compatible tool to integrate web search, validating Jan's long-standing bet on MCP as the connective tissue of the agentic internet.^[20]

April 2026: $100M Raise at $2B Valuation

On April 30, 2026, Parallel Web Systems raised $100M at a $2B valuation — a headline that trended on X. Jan noticed the trending topic and spent over 4 minutes reading the discussion thread. This represents a significant milestone for the company Jan worked at as a DevRel contractor just months earlier, and validates the bet on AI-powered web search and agent infrastructure that Jan contributed to building.^[21]

April 2026: Parallel Docs Check

On April 28, 2026, Jan briefly visited the Parallel AI docs (Getting Started / Overview) — a quick check of the developer documentation, suggesting continued engagement with the platform even during a personal rest period.^[22]

Hacker Residency (November 2025)

During Jan's Parallel AI contract period, he attended the Hacker_Residency in Da Nang, Vietnam in late October / early November 2025. Organised by Travis (@transitive_bs), the residency hosted Jan alongside Edmund Tian, Melanie Klenner, Lilian, Miles Feldstein and others. Jan described it as: "I've never been around such an incredibly talented group of people. Shipping with such urgency and passion, all with the best possible vibes."^[11] This period ended shortly before his November 2025 hospitalisation.

Parallel Web Systems Inc.

Parallel Web Systems Inc. is the formal corporate entity behind Parallel AI. As of May 2026, the entity has a Ramp corporate card — Jan received an email on May 13 about updated Ramp sign-in requirements for Parallel Web Systems Inc., indicating he still has some administrative access or account linkage despite the DevRel contract ending in November 2025.^[23]

May 2026: INDEX by Parallel Launch

On May 20, 2026, Parallel Web Systems launched INDEX — a system that indexes websites so AI agents can discover and use them, and pays web creators for AI agent usage based on Shapley value attribution. The launch trended on X. Jan explored the announcement that morning, checking the INDEX about page and the INDEX dashboard for wilmake.com — the fact that he checked his own domain first signals personal alignment with the product's value proposition.^[24]

He also browsed Parallel's documentation — specifically the ClawHub, Agent Skills, and MCP Tool Calling docs — and checked the Parallel Web Systems front-end Member of Technical Staff job opening on Ashby. No application was filed as of May 22, but the role is aligned with Jan's SDR/uithub/openpolsia background.^[24]

The INDEX product is conceptually adjacent to Jan's own work on Agent_Friendly_Web, LLMTEXT, and uithub — all of which address the question of how AI agents should discover, access, and pay for web content. Jan's RT of @p0 on May 19 — "AI agents are the web's second user. They'll use it 1000x more than humans ever have" — further frames the alignment.^[25]

parallel-llms-txt Deploy Failure (May 2026)

On May 14, 2026, Jan received a CI/CD failure notification for parallel-web/parallel-llms-txt — a GitHub Actions deploy to Cloudflare Workers that failed on the main branch. As of May 26, the deploy has been broken for twelve consecutive days with no visible fix attempts. The exact purpose of parallel-llms-txt is not publicly documented — it may be an llms.txt aggregator, LLM routing layer, or proxy service.^{[26] [27]}

See Also

• Jan Wilmake
• Paraga_Agarwal
• Code From Anywhere
• uithub
• ActionSchema
• Career Plan 2026
• Hacker_Residency
• LLMTEXT