TeXRA: AI TeX Research Assistant for VS Code

🎓 Free for Researchers! TeXRA offers a Researcher Access Program with complimentary access to budget-friendly models from OpenAI, DeepSeek, Gemini, and more—plus a hosted Orchestrator and a roster of remote specialist agents. Sign in through the Profile view to get started—no API keys required.

TeXRA is a multi-agent research assistant for VS Code. Instead of chatting with a single model, you direct an Orchestrator that delegates to a team of specialists—researchers, numericists, reviewers, formalizers, LaTeX fixers, presenters—each with their own tools, prompts, and model. The result is a coordinated lab in your editor that drafts, reviews, computes, and formalizes rigorous scientific work alongside its LaTeX, code, figures, and PRs.

See texra.ai or the full documentation for tutorials, agent recipes, and a web-based launch page.

Why TeXRA

Orchestrator-first – the Orchestrator decomposes your task, delegates to specialists in parallel, captures their outputs as diffs, and presents proposals you approve before they touch your files. Follow-ups during delegation are queued, sub-agent runs can be inspected, waited on, resumed, or terminated, and the orchestrator builds long-term memory across sessions.
Curated team presets – switch to Physicist, Mathematician, Computer Scientist (ML), or Lean Project in one click from the Multi-Agent settings tab. Each preset is a preconfigured roster of workflow and tool-use agents tuned for that discipline; you can also save your own.
A full cast of specialists – locally bundled tool-use agents include research, numerics, review, presenter, latexFixer, latexDiff, creator, lean, chat, and the Setup Wizard (setup); workflow agents include correct, polish, merge, ocr, transcribe_audio, paper2slide, and paper2poster. Signing in unlocks remote specialists— orchestrator, search, simplifier, criticize, devise, apply, generic, progressCheck, and the Lean leanOrchestrator / leanBlueprint / leanSearch / leanSimplifier line.
Tools that touch your project – tool-use agents read and edit workspace files, run shell commands, drive LaTeX builds, work with Git and GitHub PR subscriptions, and can delegate reasoning turns to the Codex CLI—each tool call gated by per-stream approval (with an optional YOLO bypass).
Live, persistent runs – the progress board streams reasoning, tool calls, sub-agent file diffs, and per-run token usage and cost for active and recently completed runs in your workspace. Saved task state can be reopened later via Show Agent Execution History, tool-use agents can be resumed via Resume Tool-Use Agent, and Pack Output into History Folder archives a finished run's outputs into the workspace's History/ directory.
Model flexibility with guardrails – mix and match per agent: OpenAI (incl. GPT-5.5 and GPT Pro), Anthropic (incl. Claude Opus 4.7), Google Gemini, DeepSeek, xAI Grok, Moonshot Kimi, Alibaba Qwen (DashScope), Zhipu GLM, MiniMax, OpenRouter, and custom endpoints—with context management, retry/backoff, parallel-tool-call limits, and cost monitoring all configurable.

Built-in Agent Teams

Team	What the team does
Physicist	Analytical derivations, numerical experiments, literature search, slide drafting, and critical review.
Mathematician	Proofs, Lean 4 formalization, research, and LaTeX correction.
Computer Scientist (ML)	Algorithm design, experiments and ablations, literature search, critical review, and reproducibility.
Lean Project	Lean 4 projects—theorem search, tactic simplification, and blueprint-driven formalization.

Switch teams from the Multi-Agent tab in Settings, or build your own roster of workflow and tool-use agents. Teams that include remote specialists (e.g. the Orchestrator, search, simplifier) require sign-in or your own API keys configured for the providers those agents use.

Quick Start

Install the extension from the VS Code Marketplace or Open VSX.
Launch the Setup Wizard. Pick whichever is easiest:
- Click the 🚀 TeXRA: Get Started pill in the status bar (shown automatically until you sign in or add an API key).
- Click Run the setup assistant agent in the Getting Started banner at the top of the TeXRA sidebar.
- Open VS Code's Welcome / Walkthroughs page and pick TeXRA: Getting Started — Step 1 is a one-click button.
- Or, from the command palette, run TeXRA: Run Setup Assistant Agent (Setup Wizard).
The Setup Wizard diagnoses your environment, installs missing LaTeX tooling, helps you sign in or add an API key, and verifies you're ready to run agents—asking before every command and explaining what it's doing. It can hand off interactive sudo prompts and installers to your VS Code terminal.
Pick an agent team in Settings → Multi-Agent (Physicist, Mathematician, CS/ML, or Lean Project), or stay with the default lineup.
Open the TeXRA sidebar, select the Orchestrator (or any agent), describe your task, and approve the proposals it routes to specialists. Watch progress, file diffs, and live reasoning on the progress board, and follow up at any time—messages are queued for whichever sub-agent needs them.

New here? Use the Create Sample Project button in the Getting Started banner (also available as TeXRA: Create Sample Project from the command palette) to spin up a fully configured workspace.

Requirements

VS Code 1.105+ (also runs in compatible editors such as Cursor, Windsurf, and Google Antigravity)
LaTeX distribution (TeX Live, MiKTeX, or MacTeX) for compilation and related tooling
Perl (required by latexindent and latexdiff)
Optional: GraphicsMagick/ImageMagick and Ghostscript for PDF and image processing; git for repository-aware features; gh and a Codex CLI for GitHub PR and Codex integrations; Lean 4 + lake for the Lean Project team

The Setup Wizard checks for and helps install most of the above for you.

Configuring Models

Sign in through the Profile view to use the Researcher Access Program (which also unlocks the hosted Orchestrator and remote specialists), or store your own API keys via the TeXRA: Set API Key command (kept in VS Code's encrypted SecretStorage), or place them in a workspace .env file:

OPENAI_API_KEY=your_openai_key_here
ANTHROPIC_API_KEY=your_anthropic_key_here
GOOGLE_API_KEY=your_google_key_here
DEEPSEEK_API_KEY=your_deepseek_key_here
XAI_API_KEY=your_xai_key_here
OPENROUTER_API_KEY=your_openrouter_key_here

Other supported providers follow the same <PROVIDER>_API_KEY convention: MOONSHOT_API_KEY, DASHSCOPE_API_KEY (Qwen), MINIMAX_API_KEY, GLM_API_KEY. TeXRA loads the .env file automatically at startup. Each agent in a team can use a different model, so you can pair a flagship reasoner for the orchestrator with cheaper, faster models for routine sub-tasks. See the installation guide and the models guide for details.

Customization

Configure agents, prompts, models, and reliability policy in VS Code settings or the unified Settings view (Memory, History, Models, Agents, Multi-Agent, Tools, Git, LaTeX tabs). The Multi-Agent tab covers team presets, parallel tool-call limits, compaction thresholds, retry/backoff, and the orchestrator kill toggle. Power users can define new workflow or tool-use agents in YAML or register new model handlers.

Support & Feedback

Report issues and feature requests on the GitHub issues page or email contact@texra.ai.

TeXRA

TeXRA.ai