-
Notifications
You must be signed in to change notification settings - Fork 1
Glossary
Quick definitions for the terms you'll see across the app and docs. Listed alphabetically.
The main LLM-driven loop that reads your messages, picks tools, and composes responses. Built on pi-mono. The coordinator is the specific agent configuration Research Copilot ships with — it wires in the research tools and skill registry. Source: lib/agents/coordinator.ts.
A structured piece of knowledge the agent can create, edit, and reference. Five types:
| Type | What it holds |
|---|---|
note |
Freeform Markdown notes |
paper |
Bibliographic records with DOI, bibtex, citeKey, relevance |
data |
Tabular or structured data |
web-content |
Snapshots of web pages the agent fetched |
tool-output |
Raw output from tool calls, preserved for reference |
Artifacts live under <workspace>/.research-pilot/artifacts/<type>/ as files. They're searchable and citable via @-mentions.
A classic ranking function used by the Paper Wiki's search. Matches documents to queries by term frequency and inverse document frequency — no embeddings, no vector DB, no network calls. Fast and offline.
The rule Paper Wiki uses to decide if two paper records refer to the same work: DOI > arxivId > normalized(title+year). Prevents the same paper from being reprocessed under different keys across projects. See Paper Wiki → Canonical identity.
See Agent / Coordinator.
A rule-based (and optionally LLM-confirmed) classification of what the user is asking for — one of literature, data, writing, critique, web, citation, grants, docx, or general. Drives which skills get surfaced to the agent. Source: lib/agents/coordinator.ts.
An inline reference to an artifact inside a chat message. Typing @ opens a picker; the agent resolves the mention and the referenced content is attached to the message. Parser and resolver live in lib/mentions/.
A global, cross-project knowledge base of papers you've touched. See Paper Wiki for the deep dive.
The agent runtime Research Copilot is built on — @mariozechner/pi-coding-agent. Handles LLM integration, session management, tool invocation, and the built-in coding toolkit (read, write, edit, bash, grep, find).
A condensed Markdown summary of a chat session, generated automatically as sessions grow. Enables the agent to recall earlier context without re-loading every message. Stored at <workspace>/.research-pilot/memory-v2/session-summaries/.
A lazy-loaded knowledge module in SKILL.md format that gives the agent domain expertise. Only the shortDescription is loaded at startup; the body loads on demand when the skill is activated. Discovered from three locations (later overrides earlier): builtin → ~/.research-pilot/skills/ → <workspace>/.pi/skills/. See the Skills Catalog.
Signing in with ChatGPT Pro / Plus or Claude Pro / Max via OAuth, instead of using an API key. No per-token billing. Tokens are stored in the OS keychain. Research Copilot prefers subscription auth when available — see Getting Started → Sign in.
A typed function the agent can call mid-conversation. Research Copilot ships with research-specific tools (literature search, web fetch, data analysis, document conversion, artifact CRUD, wiki search) plus pi-mono's general-purpose coding tools (read, write, edit, bash, grep, find). Source: lib/tools/.
The folder you opened in Research Copilot. All project-local state — artifacts, session summaries, custom skills — lives under <workspace>/.research-pilot/ or <workspace>/.pi/. Global state (API keys, Paper Wiki) lives under ~/.research-pilot/ and ~/.research-copilot/.