About φ · Pheonix Research
Research work, made explainable.
φ is the conversational intelligence layer for Pheonix Research. It helps people find the right evidence, reuse accumulated knowledge, and commission structured research products when a question needs more than a quick answer.
What is the product?
Pheonix combines a conversational assistant with a set of specialised research capabilities. φ can answer a bounded question, search the shared intelligence library, explain an existing result, or help define a larger deliverable. When the work requires a defensible research process, φ starts a workflow that runs independently while the conversation remains available.
The goal is not to make every request look like a long research project. The goal is to use the lightest reliable path for the user's objective, while preserving a deeper path for decisions that require evidence, validation, and a complete output.
What are workflows?
A workflow is a planned, repeatable research product. It is not an open-ended autonomous agent. Each workflow has a defined input contract, stages, filesystem-managed instructions, validation rules, tools, persisted outputs, and a final delivery format.
Market Report
Deep market intelligence from system reconstruction and taxonomy through facts, evidence, signals, forecast, writing, and artifact generation. This is an internal/admin workflow.
Company Profiling
Authority-checked company analysis focused on a selected market, producing a structured presentation deck with financials, strategy, developments, and SWOT.
Research Desk
Light, medium, or deep research for a user question, with a documented evidence corpus, provenance, synthesis, and downloadable output.
Deck and Dashboard Generators
Source-led presentation or interactive dashboard creation. These focus on transforming supplied material or existing knowledge into a usable visual product.
Why does serious work take time?
A trustworthy output cannot be produced by asking one model call to write from memory. The system separates collection, validation, synthesis, composition, rendering, and review. This creates more work, but it makes claims traceable and gives the product a way to detect malformed outputs, unsupported numbers, missing evidence, and visual defects.
Technical model
At a high level, the product has four boundaries:
- φ orchestrator: understands the user's objective, chooses between direct capabilities and planned workflows, maintains conversation context, and supervises active work.
- Workflow runtime: executes filesystem-declared stages, assembles each stage's packet, manages progress, retries bounded failures, persists checkpoints, and emits live events.
- Tools: perform bounded operations such as web search, database reads, calculations, document rendering, chart generation, and visual review.
- Knowledge and delivery stores: retain structured research records and provenance. Final user artifacts are packaged for download with ownership and access checks.
Stage prompts, state definitions, packet policies, and tool contracts are kept outside the core runtime wherever possible. This allows a workflow to evolve by changing its declared files rather than rewriting the orchestrator for every market or chapter.
Access, privacy, and outputs
Free access supports conversation and no-cost intelligence capabilities. Authenticated paid access enables metered research and production workflows while available credit remains. Administrative access is reserved for internal development and operational control.
Workflow outputs are associated with the account and job that created them. The application exposes only the artifacts a user is authorised to see. A completed package may contain HTML, a PowerPoint deck, an interactive dashboard, charts, images, structured data, and provenance depending on the workflow.
Frequently asked questions
No. The chat is the interface. Behind it are direct tools and planned workflows that can research, validate, calculate, render, supervise, and package outputs.
The intended pipeline does not use the model as the source of numerical truth. Numbers should come from supplied or verified evidence, deterministic computation, or explicitly labelled approximation where the task allows it. Validation and provenance make unsupported figures visible.
Yes. The workflow has its own progress stream and account-scoped job state. Additional long-running work is queued according to account policy, while simple conversation and permitted direct tools remain available.
Elapsed time and call counts vary by input and provider. Documentation labels measured historical values, approximate values, and unavailable telemetry separately.