@texra-ai/cli 0.38.2 → 0.38.4
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
# TeXRA
|
|
1
|
+
# TeXRA
|
|
2
2
|
|
|
3
3
|
[](https://marketplace.visualstudio.com/items?itemName=texra-ai.texra)
|
|
4
4
|
[](https://marketplace.visualstudio.com/items?itemName=texra-ai.texra)
|
|
@@ -9,225 +9,135 @@
|
|
|
9
9
|
[](https://www.npmjs.com/package/@texra-ai/cli)
|
|
10
10
|
[](https://www.npmjs.com/package/@texra-ai/cli)
|
|
11
11
|
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
>
|
|
18
|
-
> The relay runs on sponsor credits. If TeXRA helps your research, please
|
|
19
|
-
> consider supporting it via
|
|
20
|
-
> [GitHub Sponsors](https://github.com/sponsors/texra-ai) or
|
|
21
|
-
> [Buy Me a Coffee](https://buymeacoffee.com/texra.ai) to keep the program
|
|
22
|
-
> open for everyone.
|
|
23
|
-
|
|
24
|
-
**TeXRA is a multi-agent research assistant for theorists (Physics, Math, CS,
|
|
25
|
-
Engineering, etc.).** Instead of chatting with a single model, you direct an
|
|
26
|
-
**Orchestrator** that delegates to a team of specialists—researchers,
|
|
27
|
-
numericists, reviewers, formalizers, LaTeX fixers, presenters—each with their
|
|
28
|
-
own tools, prompts, and model. The result is a coordinated lab that drafts,
|
|
29
|
-
reviews, computes, and formalizes rigorous scientific work alongside its LaTeX,
|
|
30
|
-
code, figures, and PRs.
|
|
31
|
-
|
|
32
|
-
### One assistant, two surfaces
|
|
33
|
-
|
|
34
|
-
The same agents, account, and model providers are available wherever you work:
|
|
35
|
-
|
|
36
|
-
| | **VS Code extension** | **Terminal CLI** |
|
|
37
|
-
| ------------- | --------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------- |
|
|
38
|
-
| Best for | Interactive writing, side-by-side diffs, figure previews | Scripts, CI, remote/headless machines, the keyboard-first |
|
|
39
|
-
| Install | [Marketplace](https://marketplace.visualstudio.com/items?itemName=texra-ai.texra) / [Open VSX](https://open-vsx.org/extension/texra-ai/texra) | `npm install -g @texra-ai/cli` |
|
|
40
|
-
| Drive it with | Sidebar chat + the **progress board** | `texra chat` (TUI) and `texra run` (headless) |
|
|
41
|
-
|
|
42
|
-
Pick whichever fits the moment—both run the same teams on the same `.tex`
|
|
43
|
-
projects. The deeper walkthroughs live at
|
|
44
|
-
[texra.ai](https://texra.ai) and the
|
|
45
|
-
[full documentation](https://texra.ai/guide/).
|
|
46
|
-
|
|
47
|
-
## Why TeXRA
|
|
48
|
-
|
|
49
|
-
- **Orchestrator-first** – the **Orchestrator** decomposes your task,
|
|
50
|
-
delegates to specialists in parallel, captures their outputs as diffs, and
|
|
51
|
-
presents proposals you approve before they touch your files. Follow-ups
|
|
52
|
-
during delegation are queued, sub-agent runs can be inspected, waited on,
|
|
53
|
-
resumed, or terminated, and the orchestrator builds long-term memory across
|
|
54
|
-
sessions.
|
|
55
|
-
- **Curated team presets** – switch to **Physicist**, **Mathematician**,
|
|
56
|
-
**Computer Scientist (ML)**, or **Lean Project** in one click from the
|
|
57
|
-
Multi-Agent settings tab, or with `texra multi-agent run <preset>` in the
|
|
58
|
-
terminal. Each preset is a preconfigured roster of workflow and tool-use
|
|
59
|
-
agents tuned for that discipline; you can also save your own.
|
|
60
|
-
- **A full cast of specialists** – locally bundled tool-use agents include
|
|
61
|
-
`research`, `numerics`, `review`, `presenter`, `latexFixer`, `latexDiff`,
|
|
62
|
-
`creator`, `lean`, `chat`, and the **Setup Wizard** (`setup`); workflow
|
|
63
|
-
agents include `correct`, `polish`, `merge`, `ocr`, `transcribe_audio`,
|
|
64
|
-
`paper2slide`, and `paper2poster`. Signing in unlocks remote specialists—
|
|
65
|
-
`orchestrator`, `search`, `simplifier`, `criticize`, `devise`, `apply`,
|
|
66
|
-
`generic`, `progressCheck`, and the Lean `leanOrchestrator` /
|
|
67
|
-
`leanBlueprint` / `leanSearch` / `leanSimplifier` line.
|
|
68
|
-
- **Tools that touch your project** – tool-use agents read and edit
|
|
69
|
-
workspace files, run shell commands, drive LaTeX builds, work with Git and
|
|
70
|
-
GitHub PR subscriptions, and can delegate reasoning turns to the Codex
|
|
71
|
-
CLI—each tool call gated by per-stream approval (with an optional YOLO
|
|
72
|
-
bypass).
|
|
73
|
-
- **Live, persistent runs** – watch reasoning, tool calls, sub-agent file
|
|
74
|
-
diffs, and per-run token usage and cost stream in real time—on the VS Code
|
|
75
|
-
**progress board** or in the `texra chat` terminal UI. Each surface saves its
|
|
76
|
-
runs so you can reopen and resume them later: via **Show Agent Execution
|
|
77
|
-
History** and **Resume Tool-Use Agent** in VS Code, or `texra history` and
|
|
78
|
-
`texra resume <id>` in the terminal. Finished outputs can be archived into
|
|
79
|
-
the workspace's `History/` directory.
|
|
80
|
-
- **Odyssey mode (experimental)** – let a tool-use agent run a long task to
|
|
81
|
-
completion on its own. A configurable budget auto-pauses the run for your
|
|
82
|
-
approval before going further, and a dedicated panel shows progress so
|
|
83
|
-
you can step in at any time. Off by default; enable it with the
|
|
84
|
-
`texra.experimental.odyssey.enabled` VS Code setting.
|
|
85
|
-
- **Model flexibility with guardrails** – mix and match per agent: OpenAI
|
|
86
|
-
(incl. GPT-5.5 and GPT Pro), Anthropic (incl. Claude Opus 4.7), Google
|
|
87
|
-
Gemini, DeepSeek, xAI Grok, Moonshot Kimi, Alibaba Qwen (DashScope),
|
|
88
|
-
Zhipu GLM, MiniMax, OpenRouter, and custom endpoints—with context
|
|
89
|
-
management, retry/backoff, parallel-tool-call limits, and cost monitoring
|
|
90
|
-
all configurable.
|
|
91
|
-
|
|
92
|
-
## Built-in Agent Teams
|
|
93
|
-
|
|
94
|
-
| Team | What the team does |
|
|
95
|
-
| --------------------------- | ------------------------------------------------------------------------------------------------------ |
|
|
96
|
-
| **Physicist** | Analytical derivations, numerical experiments, literature search, slide drafting, and critical review. |
|
|
97
|
-
| **Mathematician** | Proofs, Lean 4 formalization, research, and LaTeX correction. |
|
|
98
|
-
| **Computer Scientist (ML)** | Algorithm design, experiments and ablations, literature search, critical review, and reproducibility. |
|
|
99
|
-
| **Lean Project** | Lean 4 projects—theorem search, tactic simplification, and blueprint-driven formalization. |
|
|
100
|
-
|
|
101
|
-
Switch teams from the Multi-Agent tab in Settings (or `texra multi-agent` in
|
|
102
|
-
the terminal), or build your own roster of workflow and tool-use agents. Teams
|
|
103
|
-
that include remote specialists (e.g. the
|
|
104
|
-
Orchestrator, `search`, `simplifier`) require sign-in or your own API keys
|
|
105
|
-
configured for the providers those agents use.
|
|
106
|
-
|
|
107
|
-
## Quick Start
|
|
108
|
-
|
|
109
|
-
Both surfaces share the same sign-in and agents, so set up whichever you'll
|
|
110
|
-
use—or both.
|
|
111
|
-
|
|
112
|
-
### In VS Code
|
|
113
|
-
|
|
114
|
-
1. Install the extension from the [VS Code Marketplace](https://marketplace.visualstudio.com/items?itemName=texra-ai.texra)
|
|
115
|
-
or [Open VSX](https://open-vsx.org/extension/texra-ai/texra).
|
|
116
|
-
2. Launch the **Setup Wizard**. Pick whichever is easiest:
|
|
117
|
-
- Click the **🚀 TeXRA: Get Started** pill in the status bar (shown
|
|
118
|
-
automatically until you sign in or add an API key).
|
|
119
|
-
- Click **Run the setup assistant agent** in the Getting Started banner
|
|
120
|
-
at the top of the TeXRA sidebar.
|
|
121
|
-
- Open VS Code's **Welcome / Walkthroughs** page and pick **TeXRA:
|
|
122
|
-
Getting Started** — Step 1 is a one-click button.
|
|
123
|
-
- Or, from the command palette, run
|
|
124
|
-
**`TeXRA: Run Setup Assistant Agent (Setup Wizard)`**.
|
|
125
|
-
|
|
126
|
-
The Setup Wizard diagnoses your environment, installs missing LaTeX
|
|
127
|
-
tooling, helps you sign in or add an API key, and verifies you're ready
|
|
128
|
-
to run agents—asking before every command and explaining what it's
|
|
129
|
-
doing. It can hand off interactive `sudo` prompts and installers to your
|
|
130
|
-
VS Code terminal.
|
|
131
|
-
|
|
132
|
-
3. Pick an agent **team** in Settings → Multi-Agent (Physicist,
|
|
133
|
-
Mathematician, CS/ML, or Lean Project), or stay with the default lineup.
|
|
134
|
-
4. Open the TeXRA sidebar, select the **Orchestrator** (or any agent),
|
|
135
|
-
describe your task, and approve the proposals it routes to specialists.
|
|
136
|
-
Watch progress, file diffs, and live reasoning on the **progress board**,
|
|
137
|
-
and follow up at any time—messages are queued for whichever sub-agent
|
|
138
|
-
needs them.
|
|
139
|
-
|
|
140
|
-
New here? Use the **Create Sample Project** button in the Getting Started
|
|
141
|
-
banner (also available as `TeXRA: Create Sample Project` from the command
|
|
142
|
-
palette) to spin up a fully configured workspace.
|
|
143
|
-
|
|
144
|
-
### In the terminal
|
|
145
|
-
|
|
146
|
-
Run the same agents on your `.tex` projects without an editor—ideal for
|
|
147
|
-
scripts, CI, and remote machines.
|
|
12
|
+
A LaTeX research assistant for VS Code and the terminal. Multi-agent
|
|
13
|
+
workflows for writing, reviewing, formalizing, and rendering academic
|
|
14
|
+
work — with every change returned as a diff you approve.
|
|
15
|
+
|
|
16
|
+
## Install
|
|
148
17
|
|
|
149
18
|
```sh
|
|
150
|
-
|
|
19
|
+
# VS Code (or Cursor, Windsurf, Antigravity)
|
|
20
|
+
code --install-extension texra-ai.texra
|
|
151
21
|
|
|
152
|
-
|
|
153
|
-
|
|
22
|
+
# Terminal — requires Node.js 22+
|
|
23
|
+
npm install -g @texra-ai/cli
|
|
154
24
|
```
|
|
155
25
|
|
|
156
|
-
|
|
26
|
+
Sign in with GitHub or Google for hosted access (no key management
|
|
27
|
+
required), or set `<PROVIDER>_API_KEY` to use your own credentials.
|
|
28
|
+
|
|
29
|
+
### Researcher Access Program
|
|
30
|
+
|
|
31
|
+
Academic researchers can sign in for complimentary hosted access to
|
|
32
|
+
a curated set of budget-friendly models — enough to run the
|
|
33
|
+
Orchestrator and the full roster of hosted specialists on real work,
|
|
34
|
+
without managing provider keys or paying per-token rates. Sign in
|
|
35
|
+
through the Profile view in VS Code, or `texra login` in the
|
|
36
|
+
terminal.
|
|
37
|
+
|
|
38
|
+
The program is sustained by the community. If TeXRA helps your
|
|
39
|
+
research, consider supporting it via
|
|
40
|
+
[GitHub Sponsors](https://github.com/sponsors/texra-ai) or
|
|
41
|
+
[Buy Me a Coffee](https://buymeacoffee.com/texra.ai) to keep it open
|
|
42
|
+
for everyone.
|
|
43
|
+
|
|
44
|
+
## Run
|
|
45
|
+
|
|
46
|
+
In VS Code: open a `.tex` file, click the TeXRA icon, pick
|
|
47
|
+
**Orchestrator** or another agent, type a task. The Setup Wizard runs
|
|
48
|
+
on first launch and checks your environment.
|
|
49
|
+
|
|
50
|
+
In the terminal:
|
|
157
51
|
|
|
158
52
|
```sh
|
|
159
|
-
texra
|
|
160
|
-
texra
|
|
161
|
-
texra run
|
|
53
|
+
texra chat # interactive tool-use session
|
|
54
|
+
texra run polish --input paper.tex # one-shot workflow
|
|
55
|
+
texra multi-agent run physicist # named team
|
|
162
56
|
```
|
|
163
57
|
|
|
164
|
-
|
|
165
|
-
`multi-agent` to inspect and launch teams; `texra history` and `texra resume`
|
|
166
|
-
to revisit or re-run past work; `texra init` to drop a project `.texra/config.json`.
|
|
167
|
-
For automation, every headless command takes `--output-format json|ndjson` and
|
|
168
|
-
`--approval-policy never|ask|yolo`. Run `texra --help` or see the
|
|
169
|
-
[documentation](https://texra.ai/guide/) for the full command reference.
|
|
58
|
+
Run history and agent settings are shared between both surfaces.
|
|
170
59
|
|
|
171
|
-
##
|
|
60
|
+
## Teams
|
|
61
|
+
|
|
62
|
+
Four built-in presets cover the most common research disciplines:
|
|
63
|
+
|
|
64
|
+
| Team | Built for |
|
|
65
|
+
| --------------------------- | -------------------------------------------------------------------------------- |
|
|
66
|
+
| **Physicist** | Analytical derivations, numerical experiments, literature search, slide drafting |
|
|
67
|
+
| **Mathematician** | Proofs, Lean 4 formalization, research, LaTeX correction |
|
|
68
|
+
| **Computer Scientist (ML)** | Algorithm design, experiments and ablations, literature, reproducibility |
|
|
69
|
+
| **Lean Project** | Mathlib search, tactic simplification, blueprint-driven formalization |
|
|
172
70
|
|
|
173
|
-
|
|
174
|
-
|
|
175
|
-
terminal CLI
|
|
176
|
-
- **LaTeX distribution** (TeX Live, MiKTeX, or MacTeX) for compilation and
|
|
177
|
-
related tooling
|
|
178
|
-
- **Perl** (required by `latexindent` and `latexdiff`)
|
|
179
|
-
- **Optional**: GraphicsMagick/ImageMagick and Ghostscript for PDF and image
|
|
180
|
-
processing; `git` for repository-aware features; `gh` and a Codex CLI for
|
|
181
|
-
GitHub PR and Codex integrations; Lean 4 + `lake` for the Lean Project team
|
|
71
|
+
Pick a team in **Settings → Multi-Agent**, or with `texra multi-agent
|
|
72
|
+
run <preset>`. Or define your own roster in YAML.
|
|
182
73
|
|
|
183
|
-
|
|
74
|
+
## Agents
|
|
184
75
|
|
|
185
|
-
|
|
76
|
+
**Workflow agents** write to disk and produce reviewable diffs:
|
|
77
|
+
`polish`, `correct`, `merge`, `ocr`, `transcribe_audio`, `paper2slide`,
|
|
78
|
+
`paper2poster`.
|
|
186
79
|
|
|
187
|
-
|
|
188
|
-
`
|
|
189
|
-
|
|
190
|
-
|
|
191
|
-
|
|
192
|
-
|
|
193
|
-
|
|
80
|
+
**Tool-use agents** work conversationally with file, shell, and search
|
|
81
|
+
access: `research`, `numerics`, `review`, `presenter`, `latexFixer`,
|
|
82
|
+
`latexDiff`, `creator`, `lean`, `chat`, `setup`.
|
|
83
|
+
|
|
84
|
+
**Hosted specialists** (signed-in users): `orchestrator`, `search`,
|
|
85
|
+
`simplifier`, `criticize`, `devise`, `apply`, `generic`,
|
|
86
|
+
`progressCheck`, and the Lean line — `leanOrchestrator`,
|
|
87
|
+
`leanBlueprint`, `leanSearch`, `leanSimplifier`.
|
|
88
|
+
|
|
89
|
+
Every tool call is gated by per-stream approval. Optional YOLO mode
|
|
90
|
+
skips approval for autonomous runs.
|
|
91
|
+
|
|
92
|
+
## Models
|
|
93
|
+
|
|
94
|
+
Bring your own keys for OpenAI, Anthropic, Google Gemini, DeepSeek,
|
|
95
|
+
xAI Grok, Moonshot Kimi, Alibaba Qwen, Zhipu GLM, MiniMax, OpenRouter,
|
|
96
|
+
or any OpenAI-compatible endpoint. Each agent in a team can run a
|
|
97
|
+
different model — pair a flagship reasoner for orchestration with
|
|
98
|
+
cheaper, faster models for routine sub-tasks.
|
|
99
|
+
|
|
100
|
+
In the extension, run **`TeXRA: Set API Key`** (stored in VS Code's
|
|
101
|
+
encrypted SecretStorage) or add a workspace `.env`:
|
|
194
102
|
|
|
195
103
|
```env
|
|
196
|
-
OPENAI_API_KEY
|
|
197
|
-
ANTHROPIC_API_KEY
|
|
198
|
-
GOOGLE_API_KEY
|
|
199
|
-
DEEPSEEK_API_KEY=your_deepseek_key_here
|
|
200
|
-
XAI_API_KEY=your_xai_key_here
|
|
201
|
-
OPENROUTER_API_KEY=your_openrouter_key_here
|
|
104
|
+
OPENAI_API_KEY=…
|
|
105
|
+
ANTHROPIC_API_KEY=…
|
|
106
|
+
GOOGLE_API_KEY=…
|
|
202
107
|
```
|
|
203
108
|
|
|
204
|
-
|
|
205
|
-
|
|
206
|
-
|
|
207
|
-
|
|
208
|
-
|
|
209
|
-
|
|
210
|
-
|
|
109
|
+
In the CLI, export the same variables in your shell and run with
|
|
110
|
+
`--api-mode personal`.
|
|
111
|
+
|
|
112
|
+
## Requirements
|
|
113
|
+
|
|
114
|
+
- **VS Code 1.105+** (also runs in Cursor, Windsurf, Antigravity), or
|
|
115
|
+
**Node.js 22+** for the CLI
|
|
116
|
+
- **LaTeX distribution** (TeX Live, MiKTeX, or MacTeX)
|
|
117
|
+
- **Perl** (for `latexindent` and `latexdiff`)
|
|
118
|
+
- Optional: ImageMagick + Ghostscript (for PDF/image processing),
|
|
119
|
+
`git`, `gh`, Codex CLI, Lean 4 + `lake`
|
|
120
|
+
|
|
121
|
+
The Setup Wizard checks for and helps install most of the above.
|
|
122
|
+
|
|
123
|
+
## Documentation
|
|
211
124
|
|
|
212
|
-
|
|
125
|
+
- [Installation](https://texra.ai/guide/installation)
|
|
126
|
+
- [Quick Start](https://texra.ai/guide/quick-start)
|
|
127
|
+
- [Built-in Agents](https://texra.ai/guide/built-in-agents)
|
|
128
|
+
- [Polish a draft](https://texra.ai/guide/workflows/polish-a-draft) — workflow example
|
|
129
|
+
- [Models](https://texra.ai/guide/models)
|
|
130
|
+
- [Custom Agents](https://texra.ai/guide/custom-agents)
|
|
213
131
|
|
|
214
|
-
|
|
215
|
-
or the unified Settings view (Memory, History, Models, Agents, Multi-Agent,
|
|
216
|
-
Tools, Git, LaTeX tabs). The Multi-Agent tab covers team presets, parallel
|
|
217
|
-
tool-call limits, compaction thresholds, retry/backoff, and the orchestrator
|
|
218
|
-
kill toggle. In the terminal, `texra init` writes a project
|
|
219
|
-
`.texra/config.json` with the same knobs, and any setting can be overridden
|
|
220
|
-
per run with a command-line flag. Power users on either surface can define new
|
|
221
|
-
workflow or tool-use agents in YAML or register new model handlers.
|
|
132
|
+
Full docs at [texra.ai/guide](https://texra.ai/guide/).
|
|
222
133
|
|
|
223
|
-
## Support
|
|
134
|
+
## Support
|
|
224
135
|
|
|
225
|
-
|
|
226
|
-
[
|
|
227
|
-
[contact@texra.ai](mailto:contact@texra.ai).
|
|
136
|
+
Issues and feature requests: [GitHub](https://github.com/texra-ai/texra-issues/issues).
|
|
137
|
+
Contact: [contact@texra.ai](mailto:contact@texra.ai).
|
|
228
138
|
|
|
229
139
|
## License
|
|
230
140
|
|
|
231
141
|
© TeXRA Team 2025–2026. All rights reserved.
|
|
232
142
|
|
|
233
|
-
[Terms of Service](https://texra.ai/terms) · [Provider
|
|
143
|
+
[Terms of Service](https://texra.ai/terms) · [Provider list](https://texra.ai/providers)
|