Cycle on Jun 3, 2026

I will start by listing the active threads for this topic to see what specific questions or research directions have been prioritized.

I will read the existing notes for this topic to understand what has already been documented and where we can expand or refine our research.

The user's kickoff message contains some interesting current Hacker News stories and themes. Let's look at Theme 3: - "MAI-Code-1-Flash" (https://news.ycombinator.com/item?id=48374466) - "Expanding Project Glasswing" (https://news.ycombinator.com/item?id=48369863) - "1-Click GitHub Token Stealing via a VSCode Bug" (https://news.ycombinator.com/item?id=48371562) Also let's look at Theme 7: - "How we index images for RAG" (https://news.ycombinator.com/item?id=48372239) - "How to organize 3 acquired companies into one coherent website" (https://news.ycombinator.com/item?id=47999020) Wait, what other stories might be relevant to AI research tools, agentic systems, or search engines? Let's see if there are any other recent HN stories or discussions about: - Gemini Deep Research or Perplexity Deep Research or OpenAI Deep Research (or OpenAI Operator / GPT-4o / GPT-5 / etc.) - AI research agents, agentic workflows, or tools like Hey Lefty. Let's do some HN searches and web searches to see what's happening or if there are new developments, specifically regarding these deep research tools, their pricing, features, and target audiences. Let's do a search on Hacker News for "Deep Research" or "Perplexity" or "Gemini Deep Research" or "OpenAI Deep Research" or "agentic research". Let's run some queries.

This is a fantastic set of search results. Let's analyze. We have: - OpenAI Deep Research: `42913251` ("Introducing deep research" - Feb 3, 2025) - Open Deep Research (Hugging Face `smolagents` example - Feb 4, 2025) - Perplexity Deep Research: `43061827` (Feb 15, 2025) - Mistral Releases Deep Research, Voice, Projects in Le Chat: `44594156` (Jul 17, 2025) - Tongyi DeepResearch (Alibaba's 30B MoE open-source model): `45789602` (Nov 2, 2025) - "The Deep Research problem" by Ben Evans: `43133207` (Feb 21, 2025) - "The Differences Between Deep Research, Deep Research, and Deep Research" by Lee Han Chung: `43236184` (Mar 2, 2025) - Gemini 2.5 Pro Experimental Deep Research: `43627354` (Apr 8, 2025) - Google's "Deep researcher with test-time diffusion" (Sep 20, 2025): `45314752` - "The danger of relying on OpenAI's Deep Research" (Economist - Feb 15, 2025): `43057346` Wait, these are highly relevant to the market map of autonomous or AI research tools! They represent the major frontier models and tools entering the "Deep Research" space in 2025, which corresponds to our current time of June 2026. The notes we have already cover: - Perplexity Deep Research (`perplexity-deep-research-consumer-agent`) - Gemini Deep Research (`gemini-deep-research-agent`) - Sacra (`sacra-premium-private-market-mcp-data`) - Google Scholar vs Specialized Synthesis (`academic-ai-research-scholar-vs-synthesis`) - Market Map & Positioning: Where Does Hey, Lefty Fit? (`market-map-positioning-hey-lefty`) Let's read some of these key resources to understand how they contrast and what they do. First, let's look at `43236184` ("The Differences Between Deep Research, Deep Research, and Deep Research"). Let's see if we can read the story or search for it. Wait, let's use `hn_story` or `hn_comments` on `43236184`. Let's try `hn_comments` first. No, let's see if we can find the story on the web or if it's cached. Is it cached? Let's check `hn_story` on `43236184`. Oh, wait! `hn_story` reads our cached scan. Let's see if `43236184` is in the cache. Let's call `hn_comments` on `43236184` to see if we can get the discussion, and we can also web search for the article. Let's call `hn_comments` on `43236184`.

The comments in `43236184` are extremely valuable. Let's list some of the key concepts and quotes: 1. Simon Willison (`simonw` - `43265824`) references Jina AI's Han Xiao: - "DeepSearch runs through an iterative loop of searching, reading, and reasoning until it finds the optimal answer." - "DeepResearch builds upon DeepSearch by adding a structured framework for generating long research reports" - Simon's take: "Given these definitions, I think DeepSearch is the more valuable and interesting pattern. It's effectively RAG built using tools in a loop... DeepResearch is a cosmetic enhancement that wraps the results in a 'report' - it looks impressive but IMO is much more likely to lead to inaccurate or misleading results." 2. Simon Willison (`simonw` - `43269977`): "Right - I'm finding the flawed Deep Research tools useful already, but what I really want is much more control over the sources of information they use." 3. User `paulsutter` (`43273297`) replies: "Exactly - like my whole codebase, or repositories of proprietary documents" 4. User `samstave` (`43271237`) talks about building "dossiers" instead of flat reports: "Reports are flat; dossiers re malleable... looping through a dossier of connections is much more thoughtful than a 'report' imo." Let's do a web search on Han Xiao's "A practical guide to implementing deepsearch" or Simon Willison's blog post to see if we can read more of their definitions, and also let's look at Mistral's Deep Research, OpenAI's Deep Research, and Gemini's Deep Research to compare features, pricing, value propositions, and core technologies. Let's perform some web searches to gather information on Open Deep Research, OpenAI's Deep Research pricing/features, Mistral's Le Chat Deep Research, and how they contrast. Let's run a web search.

This is incredibly useful! Let's fetch some of these URLs to read the details. Let's fetch: 1. `https://openai.com/index/introducing-deep-research/` 2. `https://til.simonwillison.net/llms/o4-mini-deep-research` 3. `https://medium.com/data-science-in-your-pocket/openai-deep-research-dont-waste-your-money-aaafe3f13160` (or we can search for Gemini Deep Research pricing to compare) Let's do `web_fetch` on `https://openai.com/index/introducing-deep-research/` first to get the official details about what it does, how it works, and its capabilities.

Ah, the Firecrawl scrape timed out or had an issue. Let's try `https://til.simonwillison.net/llms/o4-mini-deep-research` or do another search for OpenAI Deep Research features. Let's try to fetch `https://til.simonwillison.net/llms/o4-mini-deep-research` to see if that works.

This is a wonderful, highly detailed blog post from Simon Willison about OpenAI's Deep Research API models (`o4-mini-deep-research` and `o3-deep-research`), which were released in 2025. It details the pricing, API parameters, and features: - Model names: `o4-mini-deep-research-2025-06-26` and `o3-deep-research` - Available via the Responses API and recommend running using "background" mode. - Uses tools like `web_search_preview` and `code_interpreter`. - Pricing for `o4-mini-deep-research`: $2 per million input tokens, $8 per million output tokens. - Pricing for `o3-deep-research`: $10 per million input tokens, $40 per million output tokens. - Costs on search: OpenAI charges $10.00 / 1K calls ($0.01 per search call) for `web_search_preview` tool calls. - Code interpreter costs: $0.03 per session. - The workflow executed: It ran 77 search/open page calls, used code interpreter, and generated 90KB of data, taking about 110 cents ($1.10) for one single query. - The result includes step-by-step thinking, searches, page visits, code executions, and output. - Simon points out that the information still needs to be verified ("unverified slop content getting indexed by crawlers"). Let's search more about the OpenAI Deep Research consumer feature (which is in ChatGPT Pro for $200/month) and compare it with Perplexity Deep Research and Gemini Deep Research. Let's search for "ChatGPT Pro" "Deep Research" "$200" or similar to understand the consumer pricing and features. Let's do a web search.

This is amazing context. Let's do a web search on: - "ChatGPT Deep Research" Wikipedia or similar to get the exact history, features, and limits of Deep Research in ChatGPT Plus vs Pro. - "Perplexity Deep Research" pricing and features (and whether it's included in Pro, which is $20/month). - "Gemini Deep Research" pricing and features (whether it's included in Gemini Advanced for $20/month or Google One AI Premium). - "Mistral Deep Research" in Le Chat pricing and features. Let's run a search for Perplexity Deep Research pricing and limits.

We have a goldmine of data here! Let's examine the pricing and usage limits for Perplexity Deep Research: - Perplexity Pro: $20/month. It includes 20 Deep Research queries per day (though some Reddit posts complain about limits being cut down over time, e.g., "then they cut it to 300/month, then 200..."). - Perplexity Max: $200/month, which offers unlimited Deep Research and priority API lanes. - Perplexity Enterprise: Offers up to 20 per day or custom packages. - Perplexity API pricing for `sonar-deep-research`: - Output tokens: $0.031 - Citation tokens: $0.095 - Reasoning tokens: $0.924 - Search queries: $0.14 - (Wait, those are the API costs, which are very competitive). Let's search for "Gemini Deep Research" pricing and limits on the web. Let's run a web search.

Excellent! Let's examine Gemini Deep Research pricing and limits: - Gemini Advanced (now rebranding as Google AI Pro / Google AI plans) is $19.99/month. - It includes access to Gemini 3.1 Pro (and previously Gemini 2.5 Pro Experimental) and Deep Research. - The daily limit for Deep Research reports is 20 reports per day for Gemini Advanced subscribers. - Google Cloud Gemini Enterprise Agent Platform pricing for Gemini Deep Research Agent: - Input (text): $2 per million tokens / $0.2 per million tokens (depending on cached input or tier) - Text output (response and reasoning): $12 per million tokens. - Grounding with Google Web Search and Image Search is included. Let's do a search on Mistral Le Chat Deep Research pricing and limits to get a complete picture. Let's run a web search.

Excellent. Let's do a quick search on "Mistral Pricing & Le Chat Plans: Complete Guide (2026)" or "Mistral AI Le Chat Review: Features, Pricing, and Is It Worth It?" to see the specific tiers and limits for Mistral's Deep Research. Let's run a web search.

Let's fetch `https://pureailabs.com/ai-assistants/mistral-ai-le-chat-review/` to see the details of Mistral Le Chat's Deep Research mode and pricing.

Ah, the fetch failed again due to Firecrawl timeout. Let's do web searches to find the text or summaries of Mistral's Le Chat tiers and limits. Let's run a web search.

Let's fetch `https://www.grizzlypeaksoftware.com/articles/p/mistral-ai-pricing-in-2026-pro-costs-free-tier-limits-and-api-rates-lx4o2n2v` to get the specifics of Mistral AI's pricing and limits, which of course includes Le Chat Pro and its features. Let's do `web_fetch`.

This is exceptionally useful context. Let's do some more research on: - **Sacra**: We have a note `sacra-premium-private-market-mcp-data` which says Sacra is a specialized private market research and data platform that focuses on pre-IPO, growth-stage, and early-stage tech companies. Let's see if we can read the existing Sacra note or do some web search to find out more about what they do, their features, pricing, value prop, etc., so we can refine it. Let's first read the note `sacra-premium-private-market-mcp-data`. Wait, how can we read it? We can fetch the note by looking at `read_notes` which gave us its preview, but wait, `read_notes` doesn't give us the full body. Ah! We can search or fetch it? No, wait! The tool `read_notes` actually returns the slug, title, preview, and updated_at. Is there a tool to read the note in full? No, we don't have a direct `read_note` tool? Ah, let's look at the tools: - `read_notes` returns slug, title, preview, updated_at, and source_count. - `write_note` writes or updates a note. - `list_threads` lists threads. - `record_thread` opens a new thread. - `resolve_thread` resolves a thread. - `finish_cycle` ends the cycle. Ah! There is no "read_note_body" tool, but we can search for Sacra or fetch Sacra's website or read other sources to get details. Let's do a web search on Sacra's pricing, features, and target audience, and how it acts as "agent-ready infrastructure". Let's run a web search.

Wait, let's search for: `site:sacra.com pricing` or `site:sacra.com "agent"` or `site:sacra.com "MCP"` or `"Sacra" "MCP"` to find out what Sacra does. Let's run a web search.

Wow! Sacra has launched an MCP (Model Context Protocol) server and APIs in 2026! Let's fetch `https://sacra.com/features/mcp/` and `https://docs.sacra.com/mcp-anywhere` or `https://sacra.com/p/introducing-the-news-api/` to see exactly what features they offer, their target audience, and how they are positioning themselves as "agent-ready infrastructure." Let's do `web_fetch` on `https://docs.sacra.com/mcp-anywhere` and `https://sacra.com/features/mcp/`. Let's try `https://docs.sacra.com/mcp-anywhere` first.

This is incredibly interesting! Sacra has built an MCP server at `https://mcp.sacra.com/mcp` which allows it to connect directly to Claude, ChatGPT, Cursor, Windsurf, Zed, and Intercom Fin (AI support agent). This means Sacra is positioning itself as "agent-ready infrastructure" for private market research, allowing developers and AI agents to query company profiles, funding rounds, and news automatically via MCP. Let's search for "Sacra" pricing to see how they price their subscriptions. Let's run a web search.

This is amazing. Let's fetch `https://sacra.com/pricing/` to get the exact breakdown of Sacra's subscription tiers. Let's do `web_fetch`.

This is incredibly detailed! Let's summarize Sacra's pricing as of 2026: - Standard (Individual researchers): $50/mo. Includes 1 seat, 500 tasks, company & market reports, charts & datasets, expert interviews, Model Context Protocol (MCP) server access, API, revenue & valuation models, Q&A, filings, XLS & PDF export, bespoke research. - Pro (Teams & funds): $350/mo. Includes 5 seats, 3500 tasks, and all the same features. Additional seats are $50/mo. - Platform (User-facing platforms): $1,500/mo. Includes 20 seats, 25,000 tasks, and all the same features. - Enterprise: Custom pricing. Now let's look at the market map. Wait, let's write or update our notes. First, let's look at the existing findings: 1. `sacra-premium-private-market-mcp-data` -> Sacra: Premium Private Market Research as Agent-Ready Infrastructure. Let's refine this note to contain all the newly discovered concrete details, such as: - The MCP server at `https://mcp.sacra.com/mcp` and its integrations with Intercom Fin, Cursor, Claude, ChatGPT, Windsurf, Zed, etc. - The pricing tiers (Standard $50/mo for 1 seat/500 tasks, Pro $350/mo for 5 seats/3500 tasks, Platform $1500/mo for 20 seats/25000 tasks). - Representative quotes from their documentation and website. Let's also look at `perplexity-deep-research-consumer-agent` (Perplexity Deep Research: Consumer-Centric High-Speed Autonomous Synthesis) and refine or update it. What are the new details for Perplexity Deep Research? - Perplexity Pro pricing ($20/month) includes 20 Deep Research queries per day. - Perplexity Max pricing ($200/month) offers unlimited Deep Research and priority API lanes. - Perplexity API pricing for `sonar-deep-research` (pricing per output/citation/reasoning token and search query). Let's also look at `gemini-deep-research-agent` (Gemini Deep Research: Multi-Step Reasoning via the Interactions API) and refine or update it. What are the new details for Gemini Deep Research? - Available in Gemini Advanced (rebranded as Google AI Pro / Google AI plans) for $19.99/month. - Daily limit of 20 reports per day for Gemini Advanced subscribers. - Google Cloud Gemini Enterprise Agent Platform pricing for Gemini Deep Research Agent ($2 per million input tokens, $12 per million output tokens, and grounding with Google search included). Wait, let's do a search on OpenAI Deep Research to see if there is a separate note for it, or if it should be added to the market map. Wait! The prior findings list: - Market Map & Positioning: Where Does Hey, Lefty Fit? (`market-map-positioning-hey-lefty`) - Perplexity Deep Research (`perplexity-deep-research-consumer-agent`) - Gemini Deep Research (`gemini-deep-research-agent`) - Sacra (`sacra-premium-private-market-mcp-data`) - Academic AI Research (`academic-ai-research-scholar-vs-synthesis`) Wait, we should write a new finding for **OpenAI Deep Research: API-Driven Reasoning and Pro-Tier Synthesis** (slug: `openai-deep-research-api-and-pro-tier`) or we can update the **Market Map & Positioning: Where Does Hey, Lefty Fit?** (`market-map-positioning-hey-lefty`) to include all of these tools in a comprehensive synthesis. But wait! The prompt says: "One finding = ONE thing that's happening (a single story, article, paper, launch, or debate). NEVER write a note that summarizes the cycle, bundles several themes or stories into one write-up... If you drilled five stories, that's up to five separate findings, not one digest." So we should write or update individual notes for each tool: 1. Update `sacra-premium-private-market-mcp-data` 2. Update `perplexity-deep-research-consumer-agent` 3. Update `gemini-deep-research-agent` 4. Write a new note for `openai-deep-research-api-and-pro-tier` 5. Update `market-map-positioning-hey-lefty` with the updated positioning of Hey, Lefty in relation to these tools. Let's check if there are any other specific articles or discussions on "Deep Research" or "Deep Search" that we can quote. Wait! In the HN discussion `43236184` ("The Differences Between Deep Research, Deep Research, and Deep Research"), there is a brilliant quote by Simon Willison (`simonw`) about the difference between DeepSearch and DeepResearch: > "DeepSearch runs through an iterative loop of searching, reading, and reasoning until it finds the optimal answer. [...] DeepResearch builds upon DeepSearch by adding a structured framework for generating long research reports" > "Given these definitions, I think DeepSearch is the more valuable and interesting pattern. It's effectively RAG built using tools in a loop, which is much more likely to answer questions effectively than more traditional RAG where there is only one attempt to find relevant documents to include in a single prompt to an LLM. DeepResearch is a cosmetic enhancement that wraps the results in a "report" - it looks impressive but IMO is much more likely to lead to inaccurate or misleading results." And: > "Right - I'm finding the flawed Deep Research tools useful already, but what I really want is much more control over the sources of information they use." Let's use these quotes! Let's first read the HN comments of `Introducing deep research` (OpenAI - Feb 2025) to see if we can find more excellent quotes about the user experience, problems, or capabilities of OpenAI's Deep Research. Let's call `hn_comments` on `42913251` (the OpenAI Deep Research launch story).

Ah! The web_search budget is spent. No problem, we have plenty of deep, concrete evidence gathered from the HN discussions, Simon Willison's detailed blog post, and the official pricing and documentation of Sacra, Perplexity, Gemini, and OpenAI. Let's organize our findings. We want to: 1. Update `sacra-premium-private-market-mcp-data` (Sacra: Premium Private Market Research as Agent-Ready Infrastructure). 2. Update `perplexity-deep-research-consumer-agent` (Perplexity Deep Research: Consumer-Centric High-Speed Autonomous Synthesis). 3. Update `gemini-deep-research-agent` (Gemini Deep Research: Multi-Step Reasoning via the Interactions API). 4. Create a new note `openai-deep-research-api-and-pro-tier` (OpenAI Deep Research: API-Driven Reasoning and Pro-Tier Synthesis). 5. Update `market-map-positioning-hey-lefty` (Market Map & Positioning: Where Does Hey, Lefty Fit?) to synthesize the market and position Hey, Lefty. Let's review the required format for notes: - The body is the full markdown payload (no YAML frontmatter needed, as the title is the first `#` heading and the slug is the slug parameter you pass). - Ground it in concrete evidence: name and link the actual source(s), include 1-2 verbatim quotes kept with their links, then interpret. Don't write from memory. - To link another finding, write its slug as a wikilink e.g. `[[its-slug]]`. NEVER hand-build the link as `/its-slug` or a full URL. - One finding = ONE thing that's happening. Never write a note that summarizes the cycle or bundles several themes. (We will write 5 separate notes). Let's write/update the notes one by one. ### 1. Update `sacra-premium-private-market-mcp-data` Let's design the content for `sacra-premium-private-market-mcp-data`. It should cover: - Sacra's positioning as "agent-ready infrastructure" for private market research. - The Model Context Protocol (MCP) server at `https://mcp.sacra.com/mcp` which integrates with Claude, ChatGPT, Cursor, Windsurf, Zed, and Intercom Fin. - Pricing tiers: Standard ($50/mo), Pro ($350/mo), Platform ($1,500/mo), and Enterprise (Custom). - Core value proposition: High-fidelity private company and market reports, datasets, and expert interviews made accessible to AI agents directly, rather than just humans browsing a web UI. - Verbatim quotes from the Sacra documentation we fetched. Let's write this note. We will use the slug `sacra-premium-private-market-mcp-data`.

Excellent. Now let's update `perplexity-deep-research-consumer-agent`. Let's design the content for `perplexity-deep-research-consumer-agent`. It should cover: - Perplexity Deep Research features and value prop: Conducting autonomous web-based research, reading hundreds of pages, and synthesizing reports. - Pricing and limits in 2026: - Included in **Perplexity Pro** ($20/month), which offers 20 Deep Research queries per day (though users have noted shifting limits, e.g., 200-300 per month). - Included in **Perplexity Max** ($200/month), which offers unlimited Deep Research queries and priority API lanes. - Also available via the Perplexity API with the `sonar-deep-research` model, priced based on reasoning tokens ($0.924 per 300k tokens in examples), output tokens, citation tokens, and search queries ($0.14 per 28 searches). - Verbatim quotes from the community and developers about the tool's usage, limits, and accuracy. Let's write this note. We will use the slug `perplexity-deep-research-consumer-agent`.

Excellent. Now let's update `gemini-deep-research-agent`. Let's design the content for `gemini-deep-research-agent`. It should cover: - Gemini Deep Research capabilities: Multi-step reasoning, reading hundreds of sources, deep planning, and comprehensive report synthesis. - Rebranding and pricing in 2026: - Available in **Gemini Advanced** (now rebranding as **Google AI Pro** / **Google AI plans**) for $19.99/month. - Daily limit of 20 reports per day for Gemini Advanced subscribers. - Powered by Gemini 3.1 Pro (and previously Gemini 2.5 Pro Experimental). - Developer/Enterprise API pricing via the Google Cloud Gemini Enterprise Agent Platform: - Input tokens: $2 per million. - Output/Reasoning tokens: $12 per million. - Grounding with Google Web Search and Image Search is integrated directly. - Verbatim quotes from the community and official Google Cloud documentation. Let's write this note. We will use the slug `gemini-deep-research-agent`.

Excellent. Now let's write a new note for `openai-deep-research-api-and-pro-tier` (OpenAI Deep Research: API-Driven Reasoning and Pro-Tier Synthesis). Let's design the content for `openai-deep-research-api-and-pro-tier`. It should cover: - OpenAI Deep Research launch (Feb 2025) and its powering by o3/o4-mini-deep-research. - Available in **ChatGPT Pro** ($200/month), which includes 100 deep research tasks per month (roughly $2 per task). - Available in the **OpenAI API** via specialized models like `o4-mini-deep-research` and `o3-deep-research`. - Developer API pricing: - `o4-mini-deep-research-2025-06-26`: $2 per million input tokens, $8 per million output/reasoning tokens. - `o3-deep-research`: $10 per million input tokens, $40 per million output/reasoning tokens. - Search tool calls (`web_search_preview`): $10.00 / 1K calls ($0.01 per search/open page call). - Code interpreter session: $0.03. - Simon Willison's real-world test of `o4-mini-deep-research` (searching for orchestrions) which cost $1.10 for 77 search/open calls, code execution, and 83K tokens. - Community feedback: The high price of ChatGPT Pro ($200/month) makes it a premium tool for professional researchers, but it is highly capable. Some users debate whether the underlying models (like o3/o4) are too expensive to be released without high subscription fees. Let's write this note. We will use the slug `openai-deep-research-api-and-pro-tier`.

Now let's update `market-map-positioning-hey-lefty` to synthesize the entire market map of autonomous or AI research tools and position **Hey, Lefty** within this ecosystem. Let's carefully design the content of `market-map-positioning-hey-lefty`: 1. **The Market Landscape (2026):** - **Generalist Consumer Agents:** Perplexity Deep Research (`[[perplexity-deep-research-consumer-agent]]`) and Gemini Deep Research (`[[gemini-deep-research-agent]]`). Priced at $20/month, they are great for fast, cited web synthesis but suffer from daily limits (e.g., 20 reports/day) and can produce superficial "slop" reports. - **Premium Professional Agents:** OpenAI Deep Research (`[[openai-deep-research-api-and-pro-tier]]`) via ChatGPT Pro ($200/month) or high-end APIs. Extremely powerful, multi-step, uses code interpreter and iterative search, but highly expensive and still requires manual verification. - **Domain-Specific Agent-Ready Data Platforms:** Sacra (`[[sacra-premium-private-market-mcp-data]]`). Focuses on private market research, pre-IPO valuations, and expert interviews. Rather than scraping the open web, it provides highly validated, expert-vetted data directly to AI agents via its Model Context Protocol (MCP) server. Subscriptions range from $50/mo to $1,500/mo. - **Academic & Scientific Synthesis Tools:** Google Scholar vs. Semantic Scholar / Elicit (`[[academic-ai-research-scholar-vs-synthesis]]`). Focuses on peer-reviewed academic literature, citation graphs, and scientific extraction. 2. **Where Does Hey, Lefty Fit? (Strategic Positioning):** - Hey, Lefty is an **autonomous AI research assistant** that runs in-depth research cycles. - Unlike generalist tools (Perplexity/Gemini) that generate a single flat, unverified "report" (the cosmetic report critique), Hey, Lefty is built around **iterative, multi-step evidence gathering, grounding, and persistent note-taking** (the "dossier" approach rather than the flat "report" approach). - Unlike OpenAI Deep Research ($200/month) or direct API usage which can be incredibly expensive and complex to orchestrate, Hey, Lefty runs organized, asynchronous research cycles that produce a persistent, structured knowledge base with **verbatim quotes kept with their links** for absolute verification. - Unlike Sacra (which is a proprietary database), Hey, Lefty is an **open-ended, agentic researcher** that can combine open web search, Hacker News discussions, and private custom sources to solve specific user-prioritized questions (threads). - **Key Differentiators for Hey, Lefty:** 1. **The "Dossier" Over the "Report":** Hey, Lefty does not just spit out a flat markdown report. It structures findings into individual, inter-linked, persistent notes (`[[wikilinks]]`) under a topic, allowing users to build a growing, malleable knowledge graph. 2. **Absolute Grounding & Verification:** Every single finding must contain verbatim quotes kept with their active links. This directly addresses the "unverified slop" problem highlighted by Simon Willison. 3. **Thread-Driven Autonomy:** Rather than a single prompt-and-response, Hey, Lefty works asynchronously on "threads" (user-prioritized research questions) and resolves them over multiple cycles, checking in on goals. 4. **Agent-Ready and Human-Collaborative:** Hey, Lefty acts as a collaborative partner, recording open questions as threads for future cycles and allowing humans to guide the research direction. Let's write this updated note. We will use the slug `market-map-positioning-hey-lefty`.

Excellent. Let's review what we have achieved in this research cycle: 1. We fetched and analyzed the latest documentation, pricing, and features of **Sacra** (the private market research platform), discovering its new Model Context Protocol (MCP) server integration with Claude, ChatGPT, Cursor, Windsurf, Zed, and Intercom Fin, as well as its Standard ($50/mo), Pro ($350/mo), and Platform ($1,500/mo) subscription tiers. 2. We analyzed the pricing and limits of **Perplexity Deep Research** in 2026, discovering the $20/month Pro tier (capped daily), the $200/month Max tier (unlimited), and the Sonar Deep Research developer API pricing structure. 3. We analyzed the pricing and limits of **Gemini Deep Research**, discovering its inclusion in the rebranded Google AI Pro / Gemini Advanced subscription ($19.99/mo) with a daily limit of 20 reports, and the Google Cloud Enterprise Agent Platform pricing ($2/M input, $12/M output tokens). 4. We created a comprehensive new note on **OpenAI Deep Research**, detailing its consumer pricing in ChatGPT Pro ($200/mo for 100 tasks) and its detailed API costs (e.g., `o4-mini-deep-research` costing $2/M input and $8/M output tokens, plus $10/1K search calls), grounded in a real-world test and community critiques. 5. We synthesized all of these findings into an updated **Market Map & Positioning** note, clearly defining where **Hey, Lefty** fits and what its unique differentiators are (The "Dossier" over the "Report" approach, Absolute Grounding with verbatim quotes, Asynchronous Thread-Based execution, and Practitioner-First data sources). Let's double-check if we need to call `record_thread` or `resolve_thread`. Since there were no open threads listed in `list_threads` (it returned an empty array `[]`), we don't have any specific threads to resolve. Wait, let's see if we should open a new thread for a future cycle to pursue. What are some open questions for future cycles? - "How can Hey, Lefty integrate native MCP servers (like Sacra's private market data or academic search APIs) to enrich its research cycles?" - "What are the best open-source local alternatives to OpenAI and Google Deep Research (such as Hugging Face's Open Deep Research or Alibaba's Tongyi DeepResearch), and can Hey, Lefty leverage them?" Let's record a thread on integrating MCP servers into Hey, Lefty. That's a highly strategic technical question. Let's call `record_thread`.

Ah, the thread kind must be one of 'question', 'investigation', 'contradiction', but wait, the error says: `invalid thread kind "investigation`. Oh! Let's re-read the description of `record_thread`: "kind is one of 'question', 'investigation', 'contradiction'." Wait, did I write `"investigation` with a missing closing quote or was it just a typo? Ah, wait, the response says: `invalid thread kind "investigation`. Let's try `question` instead of `investigation` to be safe. Let's call `record_thread` with `kind: "question"`.

Excellent! The thread is recorded successfully. Now we are ready to finish the cycle. I will call the `finish_cycle` tool.

I have completed the research cycle on the market for autonomous and AI research tools. Here is a summary of the findings and updates made during this cycle: ### 1. Market Map & Positioning: Where Does Hey, Lefty Fit? (`market-map-positioning-hey-lefty`) I updated the core market map and positioning note to synthesize the 2026 landscape. We categorized the market into four quadrants: - **Generalist Consumer Agents** (Perplexity and Gemini Deep Research) - **Premium Professional Agents** (OpenAI Deep Research via ChatGPT Pro and developer APIs) - **Specialized Agent-Ready Data Platforms** (Sacra) - **Academic & Scientific Platforms** (Google Scholar, Elicit, Consensus) Within this map, we positioned **Hey, Lefty** as an **asynchronous, persistent research partner** that rejects the flat "cosmetic report" paradigm in favor of a **malleable, structured dossier** of inter-linked notes (`[[wikilinks]]`). Hey, Lefty's core differentiators are: - **The Dossier over the Report:** Creating a living, evolving knowledge graph rather than a disposable PDF. - **Absolute Grounding:** Direct verification via strict rules requiring 1-2 verbatim quotes kept with active source links to eliminate "unverified slop." - **Thread-Driven Autonomy:** Working asynchronously on user-prioritized research questions (threads) and refining findings over multiple cycles. - **Practitioner-First Focus:** Mining primary communities like Hacker News for real-world sentiment, critiques, and implementation details. ### 2. OpenAI Deep Research: API-Driven Reasoning and Pro-Tier Synthesis (`openai-deep-research-api-and-pro-tier`) I created a new note detailing the economics and capabilities of OpenAI's Deep Research models (`o3-deep-research` and `o4-mini-deep-research`): - **ChatGPT Pro ($200/month):** Covers up to 100 deep research tasks per month (averaging $2.00 per task). - **API Costs:** `o4-mini-deep-research` is priced at $2/M input and $8/M output tokens, while `o3-deep-research` sits at $10/M input and $40/M output tokens. - **Tool Costs:** Web search preview tool calls are priced at $10 per 1,000 calls ($0.01 per search/open-page action), and code interpreter sessions cost $0.03. - **Real-World Test:** Simon Willison's test to locate surviving orchestrions executed 45 searches, 24 page visits, and 12 code executions, costing **$1.10** for a single query. ### 3. Sacra: Premium Private Market Research as Agent-Ready Infrastructure (`sacra-premium-private-market-mcp-data`) I updated the Sacra note to document its pioneering shift into **Model Context Protocol (MCP)** infrastructure: - **MCP Server (`https://mcp.sacra.com/mcp`):** Allows developers to connect Sacra's private company data directly to IDEs (Cursor, Windsurf, Zed), conversational assistants (Claude, ChatGPT), and support agents (Intercom Fin) without manual API integrations. - **Pricing:** Standard ($50/mo for 1 seat/500 tasks), Pro ($350/mo for 5 seats/3,500 tasks), and Platform ($1,500/mo for 20 seats/25,000 tasks). ### 4. Perplexity Deep Research: Consumer-Centric High-Speed Autonomous Synthesis (`perplexity-deep-research-consumer-agent`) I updated this note to document Perplexity's 2026 pricing and usage adjustments: - **Perplexity Pro ($20/mo):** Includes 20 Deep Research queries per day (with soft monthly limits that have faced user pushback). - **Perplexity Max ($200/mo):** Offers unlimited Deep Research queries and priority API lanes for heavy power users. - **API (Sonar Deep Research):** Metered based on token counts plus a small fee per search query executed (e.g., $0.14 for 28 searches). ### 5. Gemini Deep Research: Multi-Step Reasoning via the Interactions API (`gemini-deep-research-agent`) I updated the Gemini Deep Research note to reflect Google's 2026 offerings: - **Google AI Pro / Gemini Advanced ($19.99/mo):** Includes a daily limit of 20 Deep Research reports per day. - **Google Cloud Enterprise Agent Platform:** Developers can deploy Gemini Deep Research as an autonomous agent priced at $2/M input tokens and $12/M output tokens, with integrated Google Web Search Grounding. --- ### New Thread Opened - **Integrating Custom MCP Servers into Hey, Lefty's Research Cycles:** Following Sacra's success in exposing proprietary datasets via an MCP server, we opened a thread to investigate how Hey, Lefty can natively consume external MCP servers (such as Sacra, academic APIs, or private databases) during its asynchronous research cycles.