Skip to content
Back to search

tura

Build agent that uses 80% less token and delivers better results.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

Its scripts were only partly scanned.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/Tura-AI/tura with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/Tura-AI/tura'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/tura-ai--tura/badge.svg)](/listings/tura-ai--tura)

HTML

<a href="/listings/tura-ai--tura"><img src="/api/listings/tura-ai--tura/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

NPM release

Global install (macOS, Linux, and Windows):

npm install -g tura-ai
tura

Local project install:

npm install tura-ai
npx --no-install tura

The npm package does not run a `postinstall` lifecycle script. The `tura` wrapper resolves and launches the installed platform package directly.

The same main package is also published to GitHub Packages as `@tura-ai/tura`. Configure the `@tura-ai` scope for `https://npm.pkg.github.com`, authenticate with a token that has `read:packages`, then install `@tura-ai/tura`. The unscoped `tura-ai` package on npm remains the simplest public installation.

Tura does not bundle provider credentials. On first launch, configure an LLM provider and select one of its models before sending a prompt. See [Provider setup](docs/start/providers.md#first-run-configure-an-llm-provider) for the CLI, TUI, and GUI flows.

Source checkout

Windows PowerShell:

git clone https://github.com/Tura-AI/tura.git
cd tura
.\scripts\install.ps1
tura

macOS or Linux shell:

git clone https://github.com/Tura-AI/tura.git
cd tura
./scripts/install.sh
tura

The source installer performs the complete environment setup, release build, and user PATH registration flow. Pass `-EnvironmentOnly` on PowerShell or `--environment-only` on macOS/Linux only when you intentionally want dependency setup without building or registering Tura.

Common entrypoints

| Entry | Use it for | | ------------------------------------ | ---------------------------------------------------- | | `tura` | Interactive terminal UI. | | `tura "prompt"` | Open the TUI with an initial prompt. | | `tura exec "prompt"` | Direct Rust CLI prompt runner. | | `tura run "prompt"` | Gateway-backed prompt with streaming/history. | | `tura bash`, `tura zsh`, `tura shel` | Prompt with a selected command-run shell surface. | | `tura_gateway` | Local HTTP/SSE gateway and optional web GUI serving. | | `tura_gui` | Desktop GUI workspace client. |

For OS-specific PATH requirements, executor installation, and how to register the CLI when the executable is not on PATH, read [How to start](docs/start/how-to-start.md). For command flags and modes, read [CLI parameters](docs/start/cli-parameters.md).

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 106 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: tura
  3. description: Work in the Tura agent-runtime repository. Use for Tura architecture, Rust backend, GUI/TUI, prompts, commands, providers, sessions, documentation, tests, packaging, and release work in this directory.
  4. ---
  5. # Tura Repository
  6. Tura is a Rust and TypeScript agent-runtime harness that executes model-planned command graphs while reducing repeated model round trips. Use this file to load only the repository documentation relevant to the current task.
  7. ## Start Here
  8. - Read [README.md](README.md) for the product overview and primary workflows; use [README.zh-CN.md](README.zh-CN.md) for the Chinese version.
  9. - Read [docs/README.md](docs/README.md) for the documentation entry point and [docs/SUMMARY.md](docs/SUMMARY.md) for the canonical user-documentation table of contents.
  10. - Read [ARCHITECTURE.md](ARCHITECTURE.md) before changes that cross crates, processes, apps, command routing, prompts, or persistence boundaries.
  11. - Consult [ROADMAP.md](ROADMAP.md), [docs/KNOWN_ISSUES.md](docs/KNOWN_ISSUES.md), and [CHANGELOG.md](CHANGELOG.md) when evaluating intended scope, known gaps, or released behavior.
  12. ## Route by Task
  13. ### Installation, configuration, and usage
  14. - [Overview](docs/start/overview.md)
  15. - [Install and uninstall](docs/start/install.md)
  16. - [Release packages](docs/start/release-packages.md)
  17. - [Providers](docs/start/providers.md)
  18. - [How to start](docs/start/how-to-start.md)
  19. - [CLI parameters](docs/start/cli-parameters.md)
  20. - [Settings](docs/start/settings.md)
  21. - [Sessions](docs/start/sessions.md)
  22. - [Documentation navigation](docs/start/navigation.md)
  23. - [GUI settings](docs/start/gui-settings.md)
  24. - [TUI settings](docs/start/tui-settings.md)
  25. ### Runtime concepts and user-facing behavior
  26. - [Task status](docs/core/task-status.md)
  27. - [Context management](docs/core/context-management.md)
  28. - [Runtime prompt](docs/core/runtime-prompt.md)
  29. - [Command run](docs/core/command-run.md)
  30. - [Commands](docs/core/commands.md)
  31. - [Agents](docs/core/agents.md)
  32. - [Personas](docs/core/personas.md)
  33. - [HTML rich text](docs/core/html-rich-text.md)
  34. - [Prompt style and dynamic injection](docs/core/prompt-style.md)
  35. ### Customization
  36. - [Custom providers](docs/customization/custom-providers.md)
  37. - [Custom personas](docs/customization/custom-personas.md)
  38. - [Custom agents](docs/customization/custom-agents.md)
  39. - [Custom runtime prompts](docs/customization/custom-runtime-prompt.md)
  40. - [Custom commands](docs/customization/custom-commands.md)
  41. ### Component architecture
  42. - Agents: [agents/ARCHITECTURE.md](agents/ARCHITECTURE.md)
  43. - Personas: [personas/ARCHITECTURE.md](personas/ARCHITECTURE.md)
  44. - GUI: [apps/gui/README.md](apps/gui/README.md), [apps/gui/ARCHITECTURE.md](apps/gui/ARCHITECTURE.md), and [gateway adjustment notes](apps/gui/docs/gateway-adjustments.md)
  45. - TUI: [apps/tui/README.md](apps/tui/README.md) and [apps/tui/ARCHITECTURE.md](apps/tui/ARCHITECTURE.md)
  46. - Tauri shell: [apps/tauri/README.md](apps/tauri/README.md)
  47. - Gateway: [crates/gateway/ARCHITECTURE.md](crates/gateway/ARCHITECTURE.md)
  48. - Router: [crates/router/README.md](crates/router/README.md) and [crates/router/ARCHITECTURE.md](crates/router/ARCHITECTURE.md)
  49. - Runtime: [crates/runtime/ARCHITECTURE.md](crates/runtime/ARCHITECTURE.md)
  50. - Provider layer: [crates/provider/ARCHITECTURE.md](crates/provider/ARCHITECTURE.md)
  51. - Tools: [crates/tools/ARCHITECTURE.md](crates/tools/ARCHITECTURE.md)
  52. - Session log: [crates/session_log/README.md](crates/session_log/README.md) and [crates/session_log/ARCHITECTURE.md](crates/session_log/ARCHITECTURE.md)
  53. - Paths and process hardening: [crates/path/ARCHITECTURE.md](crates/path/ARCHITECTURE.md)
  54. - Build and maintenance scripts: [scripts/ARCHITECTURE.md](scripts/ARCHITECTURE.md)
  55. ### Testing and contribution
  56. Read [tests/README.md](tests/README.md) before choosing or adding a test. Test classes are peers with different isolation, cost, and external-dependency rules:
  57. - [Business tests](tests/business/README.md)
  58. - [Backend end-to-end tests](tests/e2e/README.md)
  59. - [OS tests](tests/os_testing/README.md)
  60. - [Live tests](tests/live/README.md)
  61. - [Release tests](tests/release/README.md)
  62. - [Runtime/session equivalence gate](tests/equivalence/runtime_session/README.md)
  63. - [Behavior-test quality gate](tests/equivalence/test_quality/README.md)
  64. For contribution conventions, read [docs/contributing-guide.md](docs/contributing-guide.md). [architect.md](architect.md) records the behavior-test refactor plan and is relevant only to that migration.
  65. ### Prompt implementation sources
  66. Read [the runtime-prompt guide](docs/core/runtime-prompt.md), [the prompt-style guide](docs/core/prompt-style.md), and the owning component architecture before editing prompt text. The main checked-in prompt sources are:
  67. - Agent modes: [balanced](agents/src/balanced/prompt.md), [direct](agents/src/direct/prompt.md), and [direct text only](agents/src/direct-text-only/prompt.md)
  68. - Persona and communication styles: [Tura](personas/src/tura/prompt/persona.md), [Pidan](personas/src/pidan/prompt/persona.md), [Wonderful](personas/src/wonderful/prompt/persona.md), [communication](personas/src/communication_style/communication_style.md), and [CLI communication](personas/src/communication_style/cli_communication_style.md)
  69. - Built-in commands: [generate media](commands/generate_media/prompt.md), [read media](commands/read_media/prompt.md), and [web discover](commands/web_discover/prompt.md)
  70. - Tool commands: [apply patch](crates/tools/src/commands/apply_patch/prompt.md), [bash](crates/tools/src/commands/bash/prompt.md), [planning](crates/tools/src/commands/planning/prompt.md), [shell command](crates/tools/src/commands/shell_command/prompt.md), [task status](crates/tools/src/commands/task_status/prompt.md), and [zsh](crates/tools/src/commands/zsh/prompt.md)
  71. - Runtime manuals: [data research](crates/runtime/src/runtime_prompt/data_research/prompt.md), [debug](crates/runtime/src/runtime_prompt/debug/prompt.md), [DevOps](crates/runtime/src/runtime_prompt/devops/prompt.md), [editorial](crates/runtime/src/runtime_prompt/editorial/prompt.md), [frontend](crates/runtime/src/runtime_prompt/frontend/prompt.md), [interactive and 3D](crates/runtime/src/runtime_prompt/interactive_and_3d/prompt.md), [new build](crates/runtime/src/runtime_prompt/new_build/prompt.md), [refactoring](crates/runtime/src/runtime_prompt/refactoring/prompt.md), [visual](crates/runtime/src/runtime_prompt/visual/prompt.md), and [website](crates/runtime/src/runtime_prompt/website/prompt.md)
  72. ## Historical and Project Context
  73. - Release notes: [0.1.34](docs/changelog/0.1.34.md) and [0.1.35](docs/changelog/0.1.35.md)
  74. - Project rationale: [Why I am building Tura](docs/blog/why-i-am-building-tura.md)
  75. - Benchmark and engineering context: [MCP workflow benchmark lessons](docs/blog/what-we-learned-from-the-mcp-workflow-benchmark.md), [benchmark data and test reports](docs/blog/we-need-more-benchmark-data-and-test-reports.md), [token-saving plugin analysis](docs/blog/token-saving-plugins-the-denominator-matters.md), and [GPT-5.6 Sol Max evaluation](docs/blog/is-gpt-5-6-sol-max-worth-it.md)
  76. - Contribution overview: [How to contribute to Tura](docs/blog/how-to-contribute-to-tura.md)
  77. - Funding and incubation tracking: [funding-application-tracker.md](funding-application-tracker.md)
  78. ## Working Rules
  79. - Treat the root and component architecture documents as the source of truth for ownership and boundaries; confirm behavior in code when documentation and implementation differ.
  80. - Keep user-facing documentation aligned with behavior changes. Update both language variants when a change affects claims or instructions shared by the English and Chinese root READMEs.
  81. - Use the test class required by [tests/README.md](tests/README.md); do not collapse process-owning, live, performance, release, and local business coverage into one runner.
  82. - Do not treat Markdown under `node_modules`, generated output, logs, sessions, or captured external npm research as repository instructions.
README.md · README · 386 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <p align="center">
  2. <a href="https://turaai.net/">
  3. <img src="assets/tura/icon.svg" alt="Tura icon" width="96">
  4. </a>
  5. </p>
  6. <p align="center">
  7. <a href="https://turaai.net/"><img alt="Website" title="Tura official website" src="https://img.shields.io/badge/Website-turaai.net-40e0d0?style=flat-square&amp;labelColor=555555"></a>
  8. <a href="https://turaai.net/benchmark"><img alt="Benchmark: 8,243 turns" title="Tura benchmark: 8,243 agent turns" src="https://img.shields.io/badge/Benchmark-8%2C243_turns-9b59b6?style=flat-square&amp;labelColor=555555"></a>
  9. <a href="https://www.npmjs.com/package/tura-ai"><img alt="npm package" title="Tura npm package" src="https://img.shields.io/npm/v/tura-ai?style=flat-square&amp;logo=npm&amp;label=npm&amp;labelColor=555555&amp;color=cb3837"></a>
  10. </p>
  11. <p align="center"><strong>English</strong> | <a href="README.zh-CN.md">简体中文</a></p>
  12. <h1 align="center">Tura: 16.7% better performance, 77.5% fewer tokens. </h1>
  13. Tura is an open-source agent runtime harness that delivers better results with fewer tokens.
  14. In a ReAct session, the model must re-enter after every tool result, repeatedly carrying the system prompt and a growing context. Tura turns the same task into one runtime-managed command graph, so deterministic execution continues without another model round trip.
  15. <p align="center">
  16. <img src="assets/react-vs-tura-patch-workflow.svg" alt="Animated comparison of the ReAct and Tura runtime execution loops" width="1200">
  17. </p>
  18. <p align="center"><em>Tura executes in one turn the same workflow that a ReAct architecture needs five turns to complete.</em></p>
  19. Across 20 DeepSWE v1.1 tasks, each run three times per agent, Tura creates a substantial token-budget advantage by reducing repeated context and model round trips. You can spend that advantage in two ways. Direct turns most of it into lower cost: 77.5% fewer aggregate tokens than Codex CLI, with a comparable verifier success rate of 65.0% versus 63.3%. Balanced puts more of the saved budget back into reasoning, investigation, and verification. It reached an 80.0% success rate: 16.7 percentage points higher than Codex CLI: while still using 31.1% fewer tokens.[^debug-figure][^debug-manifests]
  20. ### Benchmark
  21. Long-horizon task [benchmarks](https://turaai.net/benchmark) are one way to look past a polished isolated prompt and see how an agent handles real work. The published comparison uses harness-based development tasks with archived prompts, per-round tool calls, token usage, patches, and verifier results.
  22. > The published artifacts compare the named Tura Balanced, Tura Direct, and Codex CLI configurations on 20 DeepSWE tasks, 5 rewrite tasks, and 2 separately reviewed design tasks. [^debug-figure]. [current test-set record](https://github.com/Tura-AI/benchmark/blob/main/doc/current-test-set-record.md). [^test-set-record]
  23. The published results do not establish equivalent quality or performance for
  24. every configured provider. Broader Anthropic/Claude, Google/Gemini,
  25. OpenAI-compatible, local-provider, UI-latency, runtime/session parsing, and
  26. cross-OS measurements remain part of the documented
  27. [roadmap](ROADMAP.md) and [known evidence gaps](docs/KNOWN_ISSUES.md).
  28. <details>
  29. <summary><strong>FULL BENCHMARK REPORT</strong></summary>
  30. <p align="center">
  31. <img src="assets/data/benchmark-agent-comparison.svg" alt="DeepSWE Debug and Rewrite Repo benchmark comparison" width="800">
  32. </p>
  33. <p align="center"><em>Harness success and aggregate token usage across 25 high-difficulty tasks, 6 agent-and-model configurations, and 270 sessions. Source and calculation notes are linked below.</em></p>
  34. </details>
  35. ### Screenshots
  36. <p align="center">
  37. <img src="assets/screenshot/gui-ci-quality-demo.svg" alt="Tura GUI" width="800">
  38. </p>
  39. <p align="center"><em>GUI page with multi-session concurrent work and HTML rich text support.</em></p>
  40. <p align="center">
  41. <img src="assets/screenshot/tui-ci-quality-demo.svg" alt="Tura TUI" width="800">
  42. </p>
  43. <p align="center"><em>TUI page with multi-session concurrent work and HTML rich text support.</em></p>
  44. The results below come from published benchmark artifacts, not an uncited aggregate. Three systems do most of the work:
  45. ## Macro CLI Command Run
  46. Most coding agents still depend on repetitive tool-calling loops: inspect, wait, patch, wait, build, wait, test, wait.
  47. _**Tool-calling coding agent:**_
  48. ```bash
  49. # Turn 1 : inspect environment
  50. rg -n "TODO|command_run|handler" crates/
  51. rg --files crates/runtime/src crates/tools/src
  52. ```
  53. ```bash
  54. # Turn 2 : apply patch
  55. *** Begin Patch
  56. *** Update File: crates/tools/src/command_run/handler.rs
  57. @@
  58. - // old command handler logic
  59. + // patched command handler logic
  60. *** End Patch
  61. ```
  62. ```bash
  63. # Turn 3 : build
  64. cargo build -p runtime
  65. ```
  66. ```bash
  67. # Turn 4 : run tests
  68. cargo test -p runtime --lib
  69. ```
  70. ```bash
  71. # Turn 5 : run lint validation
  72. cargo clippy -p runtime --all-targets
  73. ```
  74. Tura takes a different route. Instead of exposing dozens of small tools to the model, it exposes one macro tool: `command_run`. The agent can then build a multi-step execution tree and run related actions in one LLM turn.
  75. In the example below, both agents run the same commands. A normal tool-calling agent needs five LLM turns; Tura handles the sequence as one structured macro workflow. The saved work is conversational overhead, not engineering discipline.
  76. _**Tura macro CLI command:**_
  77. ```json
  78. {
  79. "name": "command_run",
  80. "arguments": {
  81. "commands": [
  82. {
  83. "step": 1,
  84. "command_type": "shell_command",
  85. "command_line": "rg -n \"TODO|command_run|handler\" crates/"
  86. },
  87. {
  88. "step": 1,
  89. "command_type": "shell_command",
  90. "command_line": "rg --files crates/runtime/src crates/tools/src"
  91. },
  92. {
  93. "step": 2,
  94. "command_type": "apply_patch",
  95. "command_line": "*** Begin Patch\n*** Update File: crates/tools/src/command_run/handler.rs\n@@\n- // old command handler logic\n+ // patched command handler logic\n*** End Patch"
  96. },
  97. {
  98. "step": 3,
  99. "command_type": "shell_command",
  100. "command_line": "cargo build -p runtime"
  101. },
  102. {
  103. "step": 4,
  104. "command_type": "shell_command",
  105. "command_line": "cargo test -p runtime --lib"
  106. },
  107. {
  108. "step": 4,
  109. "command_type": "shell_command",
  110. "command_line": "cargo clippy -p runtime --all-targets"
  111. }
  112. ]
  113. }
  114. }
  115. ```
  116. There is no ablation test proving that `command_run` alone causes Tura's lower turn and token usage. Across the full DeepSWE comparison, however, Balanced used 35.8% fewer turns and 31.1% fewer tokens than Codex CLI, while Direct used 69.1% fewer turns and 77.5% fewer tokens.[^debug-figure][^debug-manifests]
  117. ## Backward Reasoning
  118. However impressive LLMs can be, an LLM is still, at its core, a statistical induction model over text-token probabilities.
  119. For example, asking an LLM to choose among rock, paper, and scissors does not guarantee a uniform random result. If a true one-in-three distribution matters, the choice needs an external random-number source rather than an uncited assumption about model output probabilities.
  120. In coding tasks, this is often fatal.
  121. An agent is more likely to execute and generate code and logic that are statistically more common. But common code and common logic are often mediocre and under-considered.
  122. Tura uses a different strategy.
  123. During reasoning, a common agent reasons from the current state to the prompt goal. In that case, $s_1$ is the current state, and $s_n$ is the goal given by the user prompt.
  124. $$
  125. s_1 \rightarrow s_2 \rightarrow s_3 \rightarrow \cdots \rightarrow s_n
  126. $$
  127. Instead, Tura guides the LLM to statistically estimate $s_{n-1}$ first, then reason backward from the state of $s_{n-1}$ to $s_{n-2}$.
  128. In the example below, the LLM can derive the optimal strategy for playing rock-paper-scissors correctly.
  129. ```
  130. > To keep rock-paper-scissors fair and challenging,
  131. > We need unbiased play.
  132. > Each move must have a true one-in-three chance.
  133. > An LLM cannot guarantee that from text probabilities alone.
  134. > Use a random-number generator script to generate randint(1, 3)
  135. > Then map rock, paper, or scissors to the number.
  136. ```
  137. In programming tasks, this means that when an agent sees a goal like fixing a frontend bug, it is guided to reason through the full execution path, reconstruct the failure state, and identify the root cause before writing code. In the published DeepSWE comparison, Tura Balanced passed 10 more of 60 binary task verifiers than Codex CLI.
  138. On the same 20-task subset, DeepSWE’s official mini-swe-agent results show an 8% gap between GPT-5.6 SOL High and Medium reasoning, while Tura Balanced leads Codex CLI by 16.7%. This indicates that higher reasoning effort alone does not explain Tura’s advantage.[^debug-manifests][^rewrite-manifest]
  139. ## Runtime Context and Prompt Manager
  140. Skills are often just weaker prompts loaded into context.
  141. In many agent frameworks, a long-lived session keeps accumulating skill files, tool outputs, and stale task history. When the context becomes too large, the agent enters a separate compaction turn, but that compaction usually preserves only a compressed summary. Important execution details can become vague or lost.
  142. Tura treats context as part of the runtime state machine.
  143. Instead of relying on users to manually reset sessions or letting Markdown skills pile up, Tura uses `task_status`, runtime prompts, and recursive execution manuals to keep the active context scoped to the current task.
  144. Traditional skill-based agents usually keep one session running until the user starts another, load broad Markdown skills into that session, and leave them active until a reset or compaction. Tura instead ties runtime prompts to explicit task state: sessions can be renamed, refreshed, and managed automatically; task-specific manuals and CLI commands are loaded through a recursive task tree; and irrelevant context can be removed, replaced, or compacted from the CLI. The checkpoint can retain code locations, patches, tests, and task status rather than only a loose summary. In practice, that means less stale context, lower task-scoped token cost, and fewer chances for an old skill or vague summary to steer the current job.
  145. Because compaction is a CLI operation, Tura can preserve exact execution state in `task_status.compact_context`. In the published benchmark sessions, Tura moved beyond read-only inspection and resumed execution an average of 2.6 rounds after compaction, compared with an estimated 5.4 rounds for Codex.[^compact-dynamodb][^compact-wasmi-r1][^compact-wasmi-r2][^compact-wasmi-r3][^compact-eza]
  146. Tura's 2.6-round result is calculated from explicit `compact_context` events in its archived round contracts. Codex does not expose equivalent compaction events, so its 5.4-round result is estimated from points where input-token usage drops sharply, excluding identifiable media-reading boundaries.
  147. ## Install and run
  148. ### NPM release
  149. Global install (macOS, Linux, and Windows):
  150. ```bash
  151. npm install -g tura-ai
  152. tura
  153. ```
  154. Local project install:
  155. ```bash
  156. npm install tura-ai
  157. npx --no-install tura
  158. ```
  159. The npm package does not run a `postinstall` lifecycle script. The `tura`
  160. wrapper resolves and launches the installed platform package directly.
  161. The same main package is also published to GitHub Packages as `@tura-ai/tura`.
  162. Configure the `@tura-ai` scope for `https://npm.pkg.github.com`, authenticate
  163. with a token that has `read:packages`, then install `@tura-ai/tura`. The
  164. unscoped `tura-ai` package on npm remains the simplest public installation.
  165. Tura does not bundle provider credentials. On first launch, configure an LLM
  166. provider and select one of its models before sending a prompt. See
  167. [Provider setup](docs/start/providers.md#first-run-configure-an-llm-provider) for
  168. the CLI, TUI, and GUI flows.
  169. ### Source checkout
  170. Windows PowerShell:
  171. ```powershell
  172. git clone https://github.com/Tura-AI/tura.git
  173. cd tura
  174. .\scripts\install.ps1
  175. tura
  176. ```
  177. macOS or Linux shell:
  178. ```bash
  179. git clone https://github.com/Tura-AI/tura.git
  180. cd tura
  181. ./scripts/install.sh
  182. tura
  183. ```
  184. The source installer performs the complete environment setup, release build,
  185. and user PATH registration flow. Pass `-EnvironmentOnly` on PowerShell or
  186. `--environment-only` on macOS/Linux only when you intentionally want dependency
  187. setup without building or registering Tura.
  188. ### Common entrypoints
  189. | Entry | Use it for |
  190. | ------------------------------------ | ---------------------------------------------------- |
  191. | `tura` | Interactive terminal UI. |
  192. | `tura "prompt"` | Open the TUI with an initial prompt. |
  193. | `tura exec "prompt"` | Direct Rust CLI prompt runner. |
  194. | `tura run "prompt"` | Gateway-backed prompt with streaming/history. |
  195. | `tura bash`, `tura zsh`, `tura shel` | Prompt with a selected command-run shell surface. |
  196. | `tura_gateway` | Local HTTP/SSE gateway and optional web GUI serving. |
  197. | `tura_gui` | Desktop GUI workspace client. |
  198. For OS-specific PATH requirements, executor installation, and how to register the
  199. CLI when the executable is not on PATH, read
  200. [How to start](docs/start/how-to-start.md). For command flags and modes, read
  201. [CLI parameters](docs/start/cli-parameters.md).
  202. ## Documentation
  203. The GitBook-style documentation index is [docs/SUMMARY.md](docs/SUMMARY.md). The
  204. full navigation page is [docs/start/navigation.md](docs/start/navigation.md).
  205. ### Start
  206. - [Overview](docs/start/overview.md)
  207. - [Install](docs/start/install.md)
  208. - [How to start](docs/start/how-to-start.md)
  209. - [CLI parameters](docs/start/cli-parameters.md)
  210. - [Settings](docs/start/settings.md)
  211. - [Providers](docs/start/providers.md)
  212. - [Sessions](docs/start/sessions.md)
  213. - [Navigation](docs/start/navigation.md)
  214. ### Core
  215. - [Task status](docs/core/task-status.md)
  216. - [Context management](docs/core/context-management.md)
  217. - [Runtime prompt](docs/core/runtime-prompt.md)
  218. - [Command run](docs/core/command-run.md)
  219. - [Commands](docs/core/commands.md)
  220. - [Agents](docs/core/agents.md)
  221. - [Personas](docs/core/personas.md)
  222. - [Rich text](docs/core/html-rich-text.md)
  223. - [Dynamic prompt injection](docs/core/prompt-style.md)
  224. ### Architecture
  225. - [System architecture](ARCHITECTURE.md)
  226. - [Runtime / Session equivalence gate](tests/equivalence/runtime_session/README.md)
  227. - [Session DB](crates/session_log/ARCHITECTURE.md)
  228. - [Gateway](crates/gateway/ARCHITECTURE.md)
  229. - [Router](crates/router/ARCHITECTURE.md)
  230. - [Runtime](crates/runtime/ARCHITECTURE.md)
  231. - [Tool](crates/tools/ARCHITECTURE.md)
  232. - [Terminal user interface](apps/tui/ARCHITECTURE.md)
  233. - [Graphic user interface](apps/gui/ARCHITECTURE.md)
  234. ### Customization
  235. - [Custom providers](docs/customization/custom-providers.md)
  236. - [Custom personas](docs/customization/custom-personas.md)
  237. - [Custom agents](docs/customization/custom-agents.md)
  238. - [Custom runtime prompt](docs/customization/custom-runtime-prompt.md)
  239. - [Custom commands](docs/customization/custom-commands.md)
  240. ### Development
  241. - [Scripts](scripts/ARCHITECTURE.md)
  242. - [Testing](scripts/ARCHITECTURE.md#xtask-test-collection-scripts)
  243. - [Environment](docs/start/settings.md)
  244. - [Architecture](ARCHITECTURE.md)
  245. - [Benchmark methodology](https://github.com/Tura-AI/benchmark/blob/main/doc/benchmark-methodology.md)
  246. - [Current test-set evidence record](https://github.com/Tura-AI/benchmark/blob/main/doc/current-test-set-record.md)
  247. - [Benchmark artifacts](https://github.com/Tura-AI/benchmark/tree/main/results)
  248. ## Contributing and project governance
  249. Contributions should be small, reviewable, and supported by evidence at the
  250. test layer that owns the affected behavior. Choose the matching issue and pull
  251. request type rather than applying one checklist to every change.
  252. - [Contributing](.github/CONTRIBUTING.md) - start here for contribution types,
  253. development setup, test selection, and pull-request steps.
  254. - [Contribution guide](docs/contributing-guide.md) - test ownership, affected
  255. matrices, performance evidence, and artifact-sanitization rules.
  256. - [Roadmap](ROADMAP.md) - current 0.1.x stabilization priorities and the planned
  257. 0.2 task-planning workspace.
  258. - [Known issues and evidence gaps](docs/KNOWN_ISSUES.md) - open architecture,
  259. provider, benchmark, performance, and cross-OS work.
  260. - [Code of Conduct](.github/CODE_OF_CONDUCT.md) - community standards and the
  261. open agent-harness principle.
  262. - [Security policy](.github/SECURITY.md) - supported versions and private
  263. vulnerability reporting.
  264. - [Support](.github/SUPPORT.md) - where to report bugs, request features, or ask
  265. setup and usage questions.
  266. ## License
  267. Tura is licensed under AGPL-3.0-or-later. See [LICENSE](LICENSE).
  268. ## Benchmark notes and sources
  269. - [Benchmark methodology](https://github.com/Tura-AI/benchmark/blob/main/doc/benchmark-methodology.md)
  270. - [Current test-set evidence record](https://github.com/Tura-AI/benchmark/blob/main/doc/current-test-set-record.md)
  271. - [Benchmark artifacts](https://github.com/Tura-AI/benchmark/tree/main/results)
  272. [^debug-figure]: [DeepSWE and Rewrite Repo comparison figure](assets/data/benchmark-agent-comparison.svg). The figure states the task, session, verifier, turn, token, and aggregation scopes used by the README.
  273. [^test-set-record]: [`tura-benchmark` current test-set record](https://github.com/Tura-AI/benchmark/blob/main/doc/current-test-set-record.md), including direct links to all eight published design HTML artifacts and their run contracts.
  274. [^debug-manifests]: [`tura-benchmark` DeepSWE replicate 1](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r01/manifest.json), [replicate 2](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r02/manifest.json), and [replicate 3](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r03/manifest.json). Each manifest contains 20 tasks across the same three agent configurations; together they contain 180 sessions.
  275. [^rewrite-manifest]: [`tura-benchmark` GPT-5.6 Rewrite Repo canonical manifest](https://github.com/Tura-AI/benchmark/blob/main/results/rewrite/report-20260710-gpt56-sol/canonical-manifest.json). The cited totals are Tura Balanced 389/472 and Codex CLI 351/472 across 10 sessions each.
  276. [^compact-dynamodb]: [`tura-benchmark` DynamoDB round 107 compact](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r01/dynamodb-toolbox-conditional-attribute-requirements/tura-balanced/dynamodb-toolbox-conditional-attribute-requirements-tura-balanced-run-01/metadata/contracts/rounds/round-0107.json) and [round 114 first later patch](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r01/dynamodb-toolbox-conditional-attribute-requirements/tura-balanced/dynamodb-toolbox-conditional-attribute-requirements-tura-balanced-run-01/metadata/contracts/rounds/round-0114.json).
  277. [^compact-wasmi-r1]: [`tura-benchmark` Wasmi replicate 1 round 43 compact](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r01/wasmi-trap-coredumps/tura-balanced/wasmi-trap-coredumps-tura-balanced-run-01/metadata/contracts/rounds/round-0043.json) and [round 44 first non-read action](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r01/wasmi-trap-coredumps/tura-balanced/wasmi-trap-coredumps-tura-balanced-run-01/metadata/contracts/rounds/round-0044.json). The run ends at round 46 without a later patch or test action.
  278. [^compact-wasmi-r2]: [`tura-benchmark` Wasmi replicate 2 round 26 compact](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r02/wasmi-trap-coredumps/tura-balanced/wasmi-trap-coredumps-tura-balanced-run-02/metadata/contracts/rounds/round-0026.json), [round 28 first non-read action](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r02/wasmi-trap-coredumps/tura-balanced/wasmi-trap-coredumps-tura-balanced-run-02/metadata/contracts/rounds/round-0028.json), and [round 39 first later patch/test](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r02/wasmi-trap-coredumps/tura-balanced/wasmi-trap-coredumps-tura-balanced-run-02/metadata/contracts/rounds/round-0039.json).
  279. [^compact-wasmi-r3]: [`tura-benchmark` Wasmi replicate 3 round 39 compact](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r03/wasmi-trap-coredumps/tura-balanced/wasmi-trap-coredumps-tura-balanced-run-03/metadata/contracts/rounds/round-0039.json) and [round 41 first later patch/test](https://github.com/Tura-AI/benchmark/blob/main/results/debug/report-deepswe-v1.1-gpt56-sol-local-r03/wasmi-trap-coredumps/tura-balanced/wasmi-trap-coredumps-tura-balanced-run-03/metadata/contracts/rounds/round-0041.json).
  280. [^compact-eza]: [`tura-benchmark` eza round 23 compact](https://github.com/Tura-AI/benchmark/blob/main/results/rewrite/report-20260710-gpt56-sol/eza/tura-balanced/eza-tura-balanced-gpt56-sol-run-02/metadata/contracts/rounds/round-0023.json) and [round 24 first later test](https://github.com/Tura-AI/benchmark/blob/main/results/rewrite/report-20260710-gpt56-sol/eza/tura-balanced/eza-tura-balanced-gpt56-sol-run-02/metadata/contracts/rounds/round-0024.json).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.