Merge pull request #9 from NVIDIA-NeMo/chore/strip-tui

Chore/strip tui

Author: rdasilveiracabral

README.md

@@ -61,14 +61,14 @@ uv add "nooa @ git+https://github.com/NVIDIA-NeMo/labs-OO-Agents.git@main"
 ```
 
 <details>
-<summary><b>Optional sub-packages</b> — CLI + TUI, memory, evaluation pipeline</summary>
+<summary><b>Optional sub-packages</b> — CLI, memory, evaluation pipeline</summary>
 
 <br />
 
 All of these live in the same repo and are addressed with `#subdirectory=…`.
 
 ```bash
-# CLI + TUI (beta): the `nooa` command, agent REPL, trace viewer, TUI
+# CLI (beta): the `nooa` command, trace viewer, eval runner
 uv add "nooa-cli @ git+https://github.com/NVIDIA-NeMo/labs-OO-Agents.git@main#subdirectory=packages/nooa-cli"
 
 # Long-term memory subsystem (MemoryManager)

THIRD_PARTY_NOTICES.md

@@ -50,90 +50,16 @@ Attribution Statements: NVIDIA includes `click` under the BSD 3-Clause License f
 License Text(https://spdx.org/licenses/BSD-3-Clause.html)
 Source/Homepage: https://github.com/pallets/click
 
-ipython - BSD 3-Clause License
-Attribution Statements: NVIDIA includes `ipython` under the BSD 3-Clause License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/BSD-3-Clause.html)
-Source/Homepage: https://github.com/ipython/ipython
-
-mcp - MIT License
-Attribution Statements: NVIDIA includes `mcp` under the MIT License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/MIT.html)
-Source/Homepage: https://github.com/modelcontextprotocol/python-sdk
-
 nooa - Apache License 2.0
 Attribution Statements: NVIDIA includes `nooa` under the Apache License 2.0 for use with NVIDIA OO Agents.
 License Text(https://spdx.org/licenses/Apache-2.0.html)
 Source/Homepage: https://github.com/NVIDIA-NeMo/labs-OO-Agents
 
-prompt-toolkit - BSD 3-Clause License
-Attribution Statements: NVIDIA includes `prompt-toolkit` under the BSD 3-Clause License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/BSD-3-Clause.html)
-Source/Homepage: https://github.com/prompt-toolkit/python-prompt-toolkit
-
-pydantic - MIT License
-Attribution Statements: NVIDIA includes `pydantic` under the MIT License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/MIT.html)
-Source/Homepage: https://github.com/pydantic/pydantic
-
 pyyaml - MIT License
 Attribution Statements: NVIDIA includes `pyyaml` under the MIT License for use with NVIDIA OO Agents.
 License Text(https://spdx.org/licenses/MIT.html)
 Source/Homepage: https://github.com/yaml/pyyaml
 
-rich - MIT License
-Attribution Statements: NVIDIA includes `rich` under the MIT License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/MIT.html)
-Source/Homepage: https://github.com/Textualize/rich
-
----
-
-## `nooa-tui`
-
-click - BSD 3-Clause License
-Attribution Statements: NVIDIA includes `click` under the BSD 3-Clause License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/BSD-3-Clause.html)
-Source/Homepage: https://github.com/pallets/click
-
-ipython - BSD 3-Clause License
-Attribution Statements: NVIDIA includes `ipython` under the BSD 3-Clause License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/BSD-3-Clause.html)
-Source/Homepage: https://github.com/ipython/ipython
-
-mcp - MIT License
-Attribution Statements: NVIDIA includes `mcp` under the MIT License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/MIT.html)
-Source/Homepage: https://github.com/modelcontextprotocol/python-sdk
-
-nooa - Apache License 2.0
-Attribution Statements: NVIDIA includes `nooa` under the Apache License 2.0 for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/Apache-2.0.html)
-Source/Homepage: https://github.com/NVIDIA-NeMo/labs-OO-Agents
-
-nooa-cli - Apache License 2.0
-Attribution Statements: NVIDIA includes `nooa-cli` under the Apache License 2.0 for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/Apache-2.0.html)
-Source/Homepage: https://github.com/NVIDIA-NeMo/labs-OO-Agents
-
-prompt-toolkit - BSD 3-Clause License
-Attribution Statements: NVIDIA includes `prompt-toolkit` under the BSD 3-Clause License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/BSD-3-Clause.html)
-Source/Homepage: https://github.com/prompt-toolkit/python-prompt-toolkit
-
-pydantic - MIT License
-Attribution Statements: NVIDIA includes `pydantic` under the MIT License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/MIT.html)
-Source/Homepage: https://github.com/pydantic/pydantic
-
-pyyaml - MIT License
-Attribution Statements: NVIDIA includes `pyyaml` under the MIT License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/MIT.html)
-Source/Homepage: https://github.com/yaml/pyyaml
-
-rich - MIT License
-Attribution Statements: NVIDIA includes `rich` under the MIT License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/MIT.html)
-Source/Homepage: https://github.com/Textualize/rich
-
 ---
 
 ## `nemo-oo-agents-benchmarks`
@@ -369,30 +295,6 @@ Source/Homepage: https://github.com/scipy/scipy
 
 ---
 
-## `nooa-cli[web]`
-
-fastapi - MIT License
-Attribution Statements: NVIDIA includes `fastapi` under the MIT License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/MIT.html)
-Source/Homepage: https://github.com/fastapi/fastapi
-
-nooa-cli[datascience] - Apache License 2.0
-Attribution Statements: NVIDIA includes `nooa-cli[datascience]` under the Apache License 2.0 for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/Apache-2.0.html)
-Source/Homepage: https://github.com/NVIDIA-NeMo/labs-OO-Agents
-
-ptyprocess - ISC License
-Attribution Statements: NVIDIA includes `ptyprocess` under the ISC License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/ISC.html)
-Source/Homepage: https://github.com/pexpect/ptyprocess
-
-uvicorn - BSD 3-Clause License
-Attribution Statements: NVIDIA includes `uvicorn` under the BSD 3-Clause License for use with NVIDIA OO Agents.
-License Text(https://spdx.org/licenses/BSD-3-Clause.html)
-Source/Homepage: https://github.com/encode/uvicorn
-
----
-
 ## `nemo-oo-agents-benchmarks[bigcodebench]`
 
 beautifulsoup4 - MIT License

docs/guides/config-migration.md

@@ -18,60 +18,7 @@ wins). Here's everything you need to move.
 
 `NAT_CONFIG_DIR` is gone — use `NEMO_OO_USER_DIR` to relocate the user dir.
 
-## 2. TUI settings: `config.toml` → `settings.yaml`
-
-Project-only TOML becomes layered YAML (`<project>/.nooa/settings.yaml` or
-user-global `~/.config/nooa/settings.yaml`). **Hard switch — `config.toml`
-is no longer read.**
-
-**Before — `config.toml`**
-```toml
-[tui]
-model = "claude-opus-4-8"
-python = false
-vi = false
-trace = ".nooa/traces"
-mcp_auto_connect = ["maas"]
-
-[tui.mcp_servers.maas]
-url = "https://your-mcp-server.example.com/mcp"
-transport = "streamable-http"
-
-[tui.mcp_servers.maas.headers]
-Authorization = "Bearer ${MAAS_API_KEY}"
-```
-
-**After — `settings.yaml`**
-```yaml
-tui:
-  default_model: claude-opus-4-8
-  show_python: false
-  vi_mode: false
-  trace_dir: .nooa/traces
-  mcp_auto_connect: [maas]
-  mcp_servers:
-    maas:
-      url: https://your-mcp-server.example.com/mcp
-      transport: streamable-http
-      headers:
-        Authorization: "Bearer ${MAAS_API_KEY}"
-```
-
-Key renames (YAML uses the real field names):
-
-| `config.toml` | `settings.yaml` |
-|---------------|-----------------|
-| `[tui]` table | `tui:` mapping |
-| `model` | `default_model` |
-| `python` | `show_python` |
-| `vi` | `vi_mode` |
-| `trace` | `trace_dir` |
-| `libs_dirs`, `mcp_auto_connect` | *(unchanged)* |
-| `[tui.mcp_servers.<name>]` | `tui.mcp_servers.<name>:` |
-
-(`/config set model …` still works — friendly keys map to the field names.)
-
-## 3. API keys / secrets
+## 2. API keys / secrets
 
 | Before | After |
 |--------|-------|
@@ -88,14 +35,14 @@ A shell `export` still wins (non-clobber), so existing exports keep working —
 `secrets.yaml` just means you no longer *need* one. The installer writes this
 file for you.
 
-## 4. LLM aliases: `llm_config.yaml`
+## 3. LLM aliases: `llm_config.yaml`
 
 Same filename and format — just move it to the new dir:
 `~/.config/nat/oo/llm_config.yaml` → `~/.config/nooa/llm_config.yaml`
 (or `<project>/.nooa/llm_config.yaml`). `nooa config eject` regenerates
 a fresh copy in the new location.
 
-## 5. Environment variables → `NEMO_OO_` prefix
+## 4. Environment variables → `NEMO_OO_` prefix
 
 | Before | After |
 |--------|-------|

examples/README.md

@@ -476,32 +476,7 @@ uv run python examples/quickstart/10_skills.py
 
 ### MCP tools
 
-MCP (Model Context Protocol) tools let your agent call external services through a standard interface. The TUI ships MCP as a hard dependency (`self.mcp` and `/mcp` are always available). In the core library, MCP is an optional extra (`uv add 'nooa[mcp]'`).
-
-**TUI configuration** — declare MCP servers in `.nooa/settings.yaml`:
-
-```yaml
-tui:
-  mcp_auto_connect: ["maas-confluence-stg"]
-  mcp_servers:
-    maas-confluence-stg:
-      url: "https://your-mcp-server.example.com/mcp"
-      transport: "streamable-http"
-      headers:
-        Authorization: "Bearer ${MAAS_API_KEY}"
-```
-
-At startup the TUI attaches auto-connected servers to the agent, so `maas-confluence-stg` becomes `self.maas_confluence_stg`. A VS Code / Claude-style `.mcp.json` file is also supported via `mcp_file` or `--mcp-file`.
-
-Inside the TUI, servers are managed through `self.mcp` (an `MCPRegistry` skill that mirrors `self.skills`):
-
-- `self.mcp.discovered()` — configured server names
-- `await self.mcp.connect(["name"])` — open a server and attach it as `self.<name>`
-- `self.mcp.activate([...])` / `self.mcp.deactivate([...])` — control whether a connected server's tools are listed for the agent
-- `<mcp>` context block lists every server each turn under **Active**, **Connected but inactive**, or **Configured**
-- Slash commands: `/mcp list`, `/mcp connect <server>`, `/mcp disconnect <server>`
-
-**Library use** — no `self.mcp`; call the stateless `MCPManager` factory directly:
+MCP (Model Context Protocol) tools let your agent call external services through a standard interface. Install with `uv add 'nooa[mcp]'` and call the stateless `MCPManager` factory:
 
 ```python
 from nooa.mcp import MCPManager

packages/nooa-cli/README.md

@@ -1,29 +1,25 @@
 # nooa-cli
 
-CLI and TUI for [nemo-oo-agents](https://github.com/NVIDIA-NeMo/labs-OO-Agents). Ships the `nooa` command (subcommands like `nooa tui`, `nooa term`, `nooa eval`, `nooa traces`) and the agent TUI (a REPL-driven CodeAct frontend).
+CLI for [nemo-oo-agents](https://github.com/NVIDIA-NeMo/labs-OO-Agents). Ships the `nooa` command with subcommands for running evaluations, browsing traces, and managing config.
 
 ## Install
 
 ```bash
-# CLI + TUI
 uv add nooa-cli
 
 # ...with numpy/pandas/plotly/scipy/sklearn pre-loaded into the LLM REPL
 uv add "nooa-cli[datascience]"
-
-# ...with web-served TUI (uvicorn + ptyprocess + fastapi, plus datascience)
-uv add "nooa-cli[web]"
 ```
 
-`nooa-cli` automatically pulls in matching `nemo-oo-agents` (the core framework). The `[datascience]` extra adds libraries the LLM can use in REPL-generated code; the `[web]` extra adds the `nooa term` web frontend.
+`nooa-cli` automatically pulls in matching `nemo-oo-agents` (the core framework). The `[datascience]` extra adds libraries the LLM can use in REPL-generated code.
 
 ## Usage
 
 ```bash
 nooa --help
-nooa tui                  # interactive agent REPL
-nooa term                 # web terminal
+nooa start-dev            # launch the trace viewer
 nooa eval ...             # eval pipeline runner
+nooa traces ...           # inspect/manage trace files
 ```
 
 See the main repo [README](https://github.com/NVIDIA-NeMo/labs-OO-Agents/blob/main/README.md) for the framework documentation.

packages/nooa-cli/pyproject.toml

@@ -2,7 +2,7 @@
 name = "nooa-cli"
 # Version is derived from git tags at build time by uv-dynamic-versioning.
 dynamic = ["version"]
-description = "CLI and TUI for nemo-oo-agents (`nooa` command, web terminal, agent REPL)"
+description = "CLI for nemo-oo-agents (`nooa` command: eval runner, trace viewer, config)"
 license = {text = "Apache-2.0"}
 readme = "README.md"
 requires-python = ">=3.12,<3.14"
@@ -12,21 +12,8 @@ dependencies = [
     # wheels declare what they actually need, not what was co-released.
     "nooa",
     "click>=8.1.0",
-    "ipython>=9.12.0",
-    # prompt-toolkit and rich were only transitive (via ipython and litellm
-    # respectively); declare them explicitly so we don't break if those
-    # transitive paths change.
-    "prompt-toolkit>=3.0.41,<3.1.0",
-    "rich>=13.0.0",
-    # Used directly in tui/agent.py and tui/models.py
-    "pydantic>=2.5.0",
-    # CLI/TUI config files
+    # CLI config files
     "pyyaml>=6.0",
-    # MCP is a first-class TUI capability (self.mcp / MCPRegistry, /mcp command,
-    # mcp_auto_connect). It stays optional in the *core* package for issue-189
-    # glibc<2.28 hosts, but the CLI already hard-depends on pydantic, so that
-    # rationale doesn't apply here — make it a hard dep of the TUI package.
-    "mcp>=1.0.0",
 ]
 
 [project.scripts]
@@ -47,26 +34,16 @@ ast = [
     "tree-sitter-rust>=0.23.0",
 ]
 # Data-science stack pre-loaded into the agent's REPL execution namespace
-# (np, pd, px, go, scipy, sklearn). Soft-imported via try/except in
-# TUI agent REPLs work without these too;
-# they exist purely so LLM-generated CodeAct cells can use these libs
-# without first failing on a missing import.
+# (np, pd, px, go, scipy, sklearn). Soft-imported via try/except; they
+# exist purely so LLM-generated CodeAct cells can use these libs without
+# first failing on a missing import.
 datascience = [
     "numpy>=1.24.0",
     "pandas>=2.0.0",
     "plotly>=5.0.0",
     "scipy>=1.10.0",
     "scikit-learn>=1.3.0",
 ]
-# Web-served TUI (`nooa term`): uvicorn + ptyprocess + fastapi, plus
-# the data-science stack so the LLM has the same REPL capabilities as
-# in the local terminal.
-web = [
-    "nooa-cli[datascience]",
-    "uvicorn[standard]>=0.34.0",
-    "ptyprocess>=0.7.0",
-    "fastapi>=0.123.4",
-]
 
 [tool.uv.sources]
 nooa = { workspace = true }

packages/nooa-cli/src/nooa_cli/__init__.py

@@ -24,13 +24,9 @@
 CONTEXT_SETTINGS = {"help_option_names": ["-h", "--help"]}
 
 
-# Thin launcher subcommands that don't touch API keys in *this* process:
-# `term` runs a web/PTY server and re-spawns the CLI as a child (which loads
-# secrets itself via bootstrap); `completion` just emits a shell script.
-# Skipping the secrets preload for them avoids importing the heavy
-# ``nooa`` core (~1.5s) on startup — keeping `nooa term` fast and
-# its SIGINT shutdown prompt.
-_SKIP_SECRETS_PRELOAD = {"term", "completion"}
+# `completion` just emits a shell script and doesn't need the ~1.5s core
+# import cost or secrets preload.
+_SKIP_SECRETS_PRELOAD = {"completion"}
 
 
 @click.group(context_settings=CONTEXT_SETTINGS)
@@ -45,10 +41,10 @@ def oo(ctx):
     if ctx.invoked_subcommand in _SKIP_SECRETS_PRELOAD:
         return
     # Load secrets.yaml into the process env before any subcommand runs so
-    # non-TUI commands (config show, eval, …) see the same API keys the TUI
-    # does. Non-clobbering (shell env wins) and best-effort: a broken or
-    # missing secrets file must never block the CLI — but log the failure at
-    # debug so it's recoverable rather than silently swallowed.
+    # all commands (config show, eval, …) see the same API keys.
+    # Non-clobbering (shell env wins) and best-effort: a broken or missing
+    # secrets file must never block the CLI — but log the failure at debug
+    # so it's recoverable rather than silently swallowed.
     try:
         from nooa.secrets import load_secrets_into_env