Written from 46 named sources · Aug 11, 2026 PigKnuckle.ai: The Deep Researcher's How-To Guide For knowledge hounds who live in frontier models — and want one workspace that routes, verifies, and delivers finished research instead of another chat thread to babysit. Feature availability, pricing, and model access verified against pigknuckle.ai as of August 11, 2026. Confirm current terms at pigknuckle.ai/pricing before relying on specific numbers. Executive Orientation & Architecture 1.1 What PigKnuckle.ai Actually Is PigKnuckle.ai is an agentic AI platform built by Alchemy Agentic [7] that takes plain-language requests and returns finished deliverables — documents, structured data outputs, comparison tables, risk summaries — with citations, supporting files, and suggested next steps [5]. It is not a chatbot. It is not a prompt engineering playground. It is a work completion engine that internally reframes your input into a structured brief, plans a multi-step execution path, routes each step to the best-suited model available within your quality tier, checks facts against real sources, and assembles the output as a finished artifact [5]. The tagline is "Stop the Slop" — a direct rejection of the chat-thread paradigm where you spend more time prompting, re-prompting, and assembling than you do reading results [5]. The product's stated philosophy: "You describe it. We finish it." [5] 1.2 Why Deep Researchers Should Care If you currently maintain accounts across Claude, ChatGPT, Gemini, and DeepSeek — switching between them depending on the task — PigKnuckle draws on models from Anthropic, OpenAI, Google, DeepSeek, and potentially others through automated per-step routing, eliminating the need to manage multiple model subscriptions [5]. You describe the research goal; the platform's stated design intent is to decide which model handles synthesis, which handles structured extraction, and which handles cost-sensitive initial scanning [5]. The critical distinction is workflow scope. A standalone Claude, GPT, or Gemini session gives you a model response inside one conversation. You then supply the research brief, source corpus, intermediate findings, formatting requirements, and revision instructions — often repeatedly and across multiple interfaces. PigKnuckle's stated product goal is to consolidate that surrounding work into a finished result with sources attached [5]. 1.3 Architecture Overview PigKnuckle describes its pipeline as following a plan-execute pattern with five stages [5]. This aligns with established agent architecture patterns seen in frameworks like LangGraph [29][32], though the specific internal implementation is not independently verified: The reframing stage is the most powerful component for deep researchers. It takes your messy, casual, half-formed input and expands it into a complete brief with the context AI needs [5]. This means you don't need to write perfect prompts — you need to write honest descriptions of what you're trying to accomplish. The reframing phase is free — you only spend credits when you approve the plan and let PigKnuckle execute [5]. Based on analysis of shared outputs and the privacy policy, PigKnuckle routes LLM calls through OpenRouter, giving it access to models from Anthropic (Claude family), OpenAI (GPT series, o-series reasoning models), Google (Gemini), DeepSeek, and potentially others [14][43]. The OpenAI-compatible API format has become the de facto standard for multi-model routing, allowing a single endpoint to reach hundreds of models by changing only the model id [13]. 1.4 Data Handling and Privacy PigKnuckle's privacy policy, last updated June 2026, states that it collects account information (email and authentication credentials), prompt data (submitted prompts and resulting enhanced prompts), usage data (transformation history and preferences), and technical data (browser type, device information, IP address) [43]. Data is stored on Supabase (managed PostgreSQL) with Cloudflare providing infrastructure services; all data transmission is encrypted in transit using TLS, and data is encrypted at rest by infrastructure providers [43]. The policy names OpenRouter and model providers, Langfuse, Sentry, Stripe, Resend, Supabase, Cloudflare, and feature-specific web-search or media-generation providers as third-party processors [43]. Critically, PigKnuckle states it does not sell personal information and routes prompts only to model providers that do not train on submitted content [43]. Retention: account content is retained while the account is active; operational event logs are deleted within 90 days; product-analytics data within 180 days; on account deletion, personal data is removed from live systems within 30 days, subject to legal retention and backup aging [43]. ⚠️ Verification required: Confirm current retention, deletion, backup, processor, and training-use terms before uploading confidential manuscripts, unpublished results, regulated data, or proprietary documentation. The privacy policy is a necessary but not sufficient guarantee for sensitive research — researchers handling pre-publication data or regulated-industry information should seek contractual assurances. Onboarding & Context Workspace 2.1 Account Setup Visit pigknuckle.ai and create an account through the "Try it free" flow. No credit card is required for the free tier [5]. You get 15 credits per month on the Free plan, which is enough to run several real research tasks [5][46]. 2.2 Understanding the Credit Model Every finished result costs credits. The reframing and planning phase is free — you only spend credits when you approve the plan and let PigKnuckle execute [5]. Credits are consumed based on the quality tier you select for each run [46]. Plan Price Credits/mo Quality Tiers Key Features Free $0 15 Fast only Web research & grounding, 3 supporting docs/journey, shareable links Starter $19/mo 75 Fast + Optimal Email support Pro $49/mo 200 Max (all models) Templates, credit rollover (up to 75/mo) Scale $99/mo 550 Max (all models) Graph execution, unlimited code execution, rollover (up to 200/mo), early beta access, priority support & onboarding Quality tier costs per run [46]: Tier Credits per Run Best For Fast 1 Quick lookups, simple deliverables, fast iteration Optimal 2 Deeper research, higher-fidelity output — the workhorse Max 4 Top models, longest context, deepest reasoning — highest-stakes work The first 3 revisions per result are free; additional revisions cost 1 credit each [46]. Worked Credit-Cost Example A typical deep-research literature review might consume credits as follows: Step Action Credits 1 Initial research task (Optimal tier) 2 2 Revision 1 — request broader sourcing 0 (free) 3 Revision 2 — strengthen specific citations 0 (free) 4 Follow-up extraction task (Fast tier) 1 Total 3 credits A more complex multi-deliverable project on Max tier with paid revisions might cost 8–12 credits. Actual costs depend on task complexity, number of steps, and revision needs. 2.3 Setting Up Your Research Environment After account creation, configure your research workspace: Confirm your plan and active quality tiers in the account dashboard [46]. Run a small, non-sensitive research task before importing a production corpus. This validates that PigKnuckle's reframing understands your domain correctly. Inspect the result format: source presentation, citation placement, supporting-document treatment, revision behavior, and export options. Test the shareable link feature — finished results have shareable URLs (e.g., pigknuckle.ai/share/...) that you can send to collaborators [5][46]. 2.4 Writing Effective Requests: The Reframing Stage Key principle: PigKnuckle is not a chat interface. Submit research briefs, not chat prompts. The reframing stage handles role assignment, context expansion, and structural framing internally [5]. Type the way you'd describe the task to a colleague — messy, casual, half-formed is fine. No "act as a" preambles, no system prompts, no model selection instructions. Instead of: ~~"You are a senior market research analyst. Please provide a comprehensive analysis of..."~~ Write: need to understand the mid-market insurance brokerage space — who are the big players, what are margins like, is there consolidation happening PigKnuckle will reframe this into a structured brief, show you the plan, and ask if you want to execute (spending credits) [5]. 2.5 Context Constraints and Input Types PigKnuckle accepts several types of input as context for your research [46]: Input Type Available On What It Does Plain-language request All tiers The primary input — messy, casual, half-formed is fine File attachment All tiers Attach documents as context for research URL attachment All tiers Attach web pages as context Entity & context profiles All tiers Define entities and context that persist across runs Supporting documents All tiers (3 free per journey) Supplementary files generated alongside main output 2.6 What to Include in Your Request Element Example Why It Matters Goal "I need to understand the competitive landscape of API gateway providers" Determines the execution path Audience "This is for a CTO making a build-vs-buy decision" Shapes tone, depth, and format Constraints "Only consider providers with SOC 2 compliance" Filters the research scope Format "I want a comparison table with pricing, features, and trade-offs" Determines output structure Prior knowledge "I've already looked at Kong and Apigee — focus on newer entrants" Avoids redundant work Source requirements "Use only peer-reviewed sources" or "cite sources" Signals verification priority 2.7 What NOT to Do Don't write system prompts or "act as a" preambles. The reframing stage handles role assignment and context expansion internally [5]. Don't over-specify the model. Let the router do its job. If you're on Pro or Scale, all models are available and the router will pick the appropriate one per step [5]. Don't write a single mega-prompt. If you have multiple distinct research questions, submit them as separate tasks — each gets its own optimized routing [5]. Dynamic Model Switching 3.1 How Routing Works PigKnuckle's core differentiator is automatic per-step model routing [5]. Rather than you choosing "I'll use Claude for this" or "Let me try Gemini," the platform's stated design intent is to: Decompose your request into discrete steps (research, extract, synthesize, format, verify) [5]. Match each step to the model best suited for it based on the quality tier you have access to [5][46]. Execute steps in sequence or parallel as appropriate. Fact-check outputs against real web sources [5]. The routing layer uses OpenRouter as the model routing infrastructure, giving access to models from Anthropic, OpenAI, Google, DeepSeek, and potentially others [14][43]. The OpenAI-compatible API format allows switching between providers by changing only the model id, making multi-model routing practical at scale [13]. 3.2 Quality Tiers and Model Access PigKnuckle exposes three quality tiers that determine which models are available for routing [46]: Tier Credits/Run Available On What It Means Fast 1 All plans (including Free) Quick, cost-efficient models for everyday work and fast iteration Optimal 2 Starter and above Deeper research and reasoning, higher-fidelity output — the workhorse Max 4 Pro and above Top models, longest context, deepest reasoning — for highest-stakes work 3.3 How to Think About Tiers as a Researcher Free tier (Fast only): Use for quick exploratory research, testing whether PigKnuckle's reframing understands your domain, and simple deliverables. 15 credits/mo is enough for approximately 15 complete tasks at Fast tier, assuming no paid revisions [46]. Starter (Fast + Optimal): The sweet spot for individual researchers who need reliable quality on most steps [46]. Pro (Max — all models): For researchers who need the strongest available models on every step. This is where multi-model routing has the most room to optimize — the router can distribute work across the full model pool for a single task [5][46]. Scale (Max + Graph + Code): For researchers running complex, multi-step workflows that require code execution (data analysis, statistical modeling) and graph-based execution paths (branching logic, parallel exploration) [5][46]. 3.4 Influencing Routing Indirectly Since the guide found no evidence of manual model selection (routing appears to be automatic per-step), you influence routing through task description. The following are suggested phrasing heuristics that may influence how the router distributes work — they are not documented routing triggers: If you want... Try describing your task as... The router may favor... Deep reasoning "Analyze the trade-offs and provide a nuanced recommendation" Stronger reasoning models (Max tier) Structured data extraction "Extract all pricing tiers into a comparison table" Models with strong structured output (Optimal/Max tier) Broad scanning "Survey the landscape and identify all major players" Cost-efficient models (Fast/Optimal tier) Code analysis "Review the code and identify architectural risks" Models with code capabilities (Max tier) Cost-efficient first pass "Quick overview of..." Fast tier models 3.5 When to Upgrade Tiers If you notice your results are technically correct but lack depth or nuance, consider these possible explanations before upgrading: Your task description may be too vague or under-specified The research domain may have limited web sources available The request may need decomposition into smaller, more focused tasks You may be hitting the ceiling of your tier's model pool Upgrading from Starter (Fast + Optimal) to Pro (Max — all models) expands the model pool available to the router, giving it access to stronger reasoning models for synthesis steps [46]. If your research involves quantitative analysis — market sizing, statistical comparisons, data transformations — the Scale tier's code execution enables quantitative analysis steps within the research workflow [5][46]. End-to-End Deep Research Workflow 4.1 The Complete Research Pipeline 4.2 Step-by-Step Execution Step 1: Frame the Research Contract Write your initial request as a deliverable specification, not a conversational prompt. Define: The research question and scope Databases or web domains to search Inclusion and exclusion criteria Required evidence fields The intended audience The output format Rules for uncertainty, conflicting findings, and unsupported claims PigKnuckle's public positioning supports casual or incomplete initial prompts being transformed into more structured work, and its examples show complete outputs built from ordinary-language requests [5]. For a scholarly review, supply the methodological constraints yourself rather than relying on the system to infer them correctly. Step 2: Review the Reframed Brief Before execution, PigKnuckle shows you what it understood you to mean — including context you didn't explicitly mention but that a domain expert would know to include [5]. Read this carefully. This is your last free checkpoint. If the reframing is wrong, edit and resubmit before spending credits. Step 3: Approve and Execute Once you approve, PigKnuckle runs the plan. PigKnuckle's marketing materials cite completion times in the range of tens of seconds for simple tasks — one shared example shows a 4-step task completing in approximately 37 seconds on the free tier [5]. More complex research tasks with more steps will take longer. Actual times depend on task complexity, model load, web research latency, and tier. Step 4: Run Extraction Before Synthesis 📋 Research methodology recommendation (not a PigKnuckle-specific feature): Ask for a structured evidence table before requesting prose. This intermediate extraction stage is a research-control measure — it makes it easier to detect that a polished conclusion has outrun the source material. Recommended columns include citation key, publication date, task, dataset, method, baseline, evaluation metric, principal result, limitation, and exact supporting passage. Step 5: Compare and Adjudicate 📋 Research methodology recommendation: Request a second pass that groups findings by evaluation design rather than by publication order. Instruct the system to distinguish: Improvements attributable to retrieval from improvements attributable to model scale Evidence from closed-book and retrieved settings Benchmark performance from production performance Statistical significance from practical significance Direct findings from author speculation Step 6: Produce the Final Artifact Require a deliverable with an executive conclusion, methods, search scope, evidence table, thematic synthesis, disagreement register, limitations, and bibliography. PigKnuckle publicly presents finished research briefings with key findings, cited sources, and a bottom line [5]. Shared output examples show source counts and reading durations (e.g., "Read 15 sources · 4m 20s") [14]. 4.3 Practical Playbooks Playbook 1: Competitive Intelligence Deep Dive Input: need a full picture of the API gateway market — who are the top 15 players, what are their pricing models, what are the real trade-offs between them, and which ones are gaining or losing momentum. want a comparison table and a memo I can share with my team What PigKnuckle does (based on shared output patterns [8][14]): Reframes into a structured market research brief (free) Routes initial web scanning to a cost-efficient model (Fast tier) Routes competitive analysis and synthesis to a reasoning model (Optimal/Max tier) Fact-checks pricing and feature claims against vendor websites Assembles a comparison table + memo with citations Suggests follow-up: "Want me to build a scoring matrix weighted to your specific criteria?" Playbook 2: Literature Review with Source Verification Input: pull together what's known about the protein leverage hypothesis — original research, meta-analyses, criticisms. I want to know if the evidence actually supports it or if it's just a compelling story. cite sources What PigKnuckle does: Reframes into a structured literature review brief Routes search and extraction steps to appropriate models Checks claims against real sources (not just generating plausible-sounding citations) [5] Assembles a structured review with inline citations Flags areas where evidence is contested or thin Suggests follow-up: "Want me to build a reference list formatted for a specific citation style?" Playbook 3: Decision Support Document Input: trying to decide whether to renew our USI insurance broker relationship or shop around. mid-market company, ~200 employees, currently paying about $1.2M/yr in premiums. need a structured analysis of what to ask, what to compare, and what the risks are of switching What PigKnuckle does (this mirrors an actual shared output — "USI Insurance Services: 2026 Mid-Market Displacement Intelligence Package" [8]): Reframes into a decision-support framework Routes market research, risk analysis, and recommendation synthesis to different models Assembles a structured intelligence package with citations — including executive summary, lead generation list, competitor intelligence brief, and territory heatmap [8] Delivers a memo, comparison framework, and risk summary Suggests follow-up: "Want me to draft the RFP for alternative brokers?" Playbook 4: Technical Infrastructure Research Input: need a brief on Cloudflare Workers ecosystem updates — Workflows V2, AI Gateway, Durable Objects changes. what's GA, what's beta, what matters for an agentic AI stack running on OpenRouter What PigKnuckle does (this mirrors an actual shared output — "Cloudflare Workers Infrastructure Update Brief" [14]): Reframes into a technical brief structure Routes technical documentation scanning and synthesis Verifies GA dates and feature claims against Cloudflare's actual docs Assembles a structured brief with citations and numbered references — including grouped release briefing, quick-reference matrix, and top 5 recommendations [14] Includes code examples and architectural recommendations Suggests follow-up: "Want me to build a migration plan from DO alarms to Workflows V2?" 4.4 Source Verification and Citation Integrity PigKnuckle explicitly claims to "check the facts against real sources" — this is a core differentiator from standard chat-based AI tools [5]. The verification stage is critical for deep research: How to maximize citation quality: Explicitly request citations. While PigKnuckle attaches sources by default, saying "cite sources" or "include references" in your request signals to the system that source verification is a priority for this task [5]. Specify source quality requirements. Saying "use only peer-reviewed sources" or "prioritize primary sources over secondary reporting" helps the system constrain its search and verification scope. Review citations in the output. PigKnuckle's shared outputs show numbered citations with source URLs [8][14][45]. Always verify that citations actually support the claims they're attached to — AI can still hallucinate plausible-looking citations. Use revisions for citation gaps. If a claim lacks a citation or the citation seems weak, request a revision (free for first 3 per result) asking for stronger sourcing on specific claims [46]. Red flags to watch for: Signal What It Might Mean Action Citation URL doesn't load Source may be hallucinated or paywalled Request revision with alternative source Citation exists but doesn't support the claim Model paraphrased incorrectly Request revision with exact quote All citations from same source Research was shallow Request broader sourcing No citations on a factual claim Verification step may have been skipped Explicitly request "every factual claim needs a citation" ⚠️ Do not accept "sources attached" as equivalent to citation validation. Open the cited documents, verify the relevant passage, and check dates, versions, authorship, and retraction or supersession status. The researcher remains responsible for checking whether each source supports the proposition attributed to it. Power-User Optimization 5.1 Advanced Workflow Patterns 📋 The following are recommended research workflow patterns proposed for this guide. They are not documented PigKnuckle features but are designed to leverage the platform's capabilities effectively. Staged Prompts for Complex Research For complex research, use staged prompts rather than a single mega-prompt: Define the research contract — scope, constraints, format Extract evidence — request a structured evidence table first Identify contradictions and gaps — ask for a disagreement register Perform adversarial review — ask the system to critique its own findings Synthesize — request the final thematic synthesis Render the final deliverable — request the formatted output Each stage is a separate credit-consuming run, but the staged approach gives you checkpoints to verify quality before investing in synthesis. The Keepsake Guide Pattern If the output is intended for long-term reference, structure it as a Keepsake Guide: Scope Provenance (source acquisition dates, access notes) Assumptions Evidence ledger Decision history Unresolved questions Dated revision record This makes the artifact reusable when sources, model behavior, or technical documentation changes. 5.2 Context Management Entity & Context Profiles PigKnuckle offers entity and context profiles on all tiers [46]. Use these to: Define recurring entities (companies, people, technologies) that appear across multiple research tasks Set persistent context constraints (e.g., "always prioritize primary sources") Reduce redundant context-setting in each new request Note: Detailed configuration guidance for entity and context profiles is not yet documented in available sources. Experiment with the feature directly. File and URL Attachment All tiers support file and U