docs: Ensure wheel readme files are suitable for pypi.org

adapters/claude/README.md

@@ -3,7 +3,7 @@ SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All
 SPDX-License-Identifier: Apache-2.0
 -->
 
-# Claude Adapter
+# NVIDIA NeMo Fabric Claude Adapter
 
 The `nvidia.fabric.claude` adapter uses the official Claude Agent SDK for
 Python behind Fabric's normalized invocation contract. The SDK is an
@@ -15,9 +15,16 @@ another executable.
 
 ## Install
 
+To install just the Claude adapter by itself:
+
+```bash
+pip install "nemo-fabric[claude]"
+```
+
+To install just the Claude adapter along with the NeMo Fabric Runtime:
+
 ```bash
-just wheels
-python -m pip install --find-links dist "nemo-fabric[claude]"
+pip install "nemo-fabric[claude, runtime]"
 ```
 
 ## Authentication
@@ -207,22 +214,3 @@ assert first.output["session_id"] == second.output["session_id"]
 Resume requires the same workspace and Claude state directory on the same host.
 The Fabric-to-Claude correlation record alone is insufficient if Claude's
 underlying transcript store is removed.
-
-## Tests
-
-The default suite uses deterministic mock Claude Code and Relay CLIs and
-requires no credentials. Test a current `nemo-relay` CLI with the mock Claude
-client, or run the live integrations on an authenticated developer host:
-
-```bash
-FABRIC_NEMO_RELAY_COMMAND="$(command -v nemo-relay)" uv run --no-sync pytest tests/e2e/test_claude.py -q -k real_relay_gateway
-RUN_FABRIC_CLAUDE_INTEGRATION=1 uv run --no-sync pytest tests/e2e/test_claude.py -q -k live
-RUN_FABRIC_CLAUDE_RELAY_INTEGRATION=1 uv run --no-sync pytest tests/e2e/test_claude.py -q -k live_claude_relay
-```
-
-The first command uses the mock Claude client and does not require credentials.
-Set `FABRIC_TEST_CLAUDE_MODEL` to override the default live-test model,
-`claude-sonnet-4-5`.
-The live Relay test applies the same semantic artifact contract as Codex: ATOF
-must contain structured LLM requests and token usage, and ATIF must contain the
-expected agent response.

adapters/claude/testing.md

@@ -0,0 +1,23 @@
+<!--
+SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+SPDX-License-Identifier: Apache-2.0
+-->
+
+# Claude Adapter Tests
+
+The default suite uses deterministic mock Claude Code and Relay CLIs and
+requires no credentials. Test a current `nemo-relay` CLI with the mock Claude
+client, or run the live integrations on an authenticated developer host:
+
+```bash
+FABRIC_NEMO_RELAY_COMMAND="$(command -v nemo-relay)" uv run --no-sync pytest tests/e2e/test_claude.py -q -k real_relay_gateway
+RUN_FABRIC_CLAUDE_INTEGRATION=1 uv run --no-sync pytest tests/e2e/test_claude.py -q -k live
+RUN_FABRIC_CLAUDE_RELAY_INTEGRATION=1 uv run --no-sync pytest tests/e2e/test_claude.py -q -k live_claude_relay
+```
+
+The first command uses the mock Claude client and does not require credentials.
+Set `FABRIC_TEST_CLAUDE_MODEL` to override the default live-test model,
+`claude-sonnet-4-5`.
+The live Relay test applies the same semantic artifact contract as Codex: ATOF
+must contain structured LLM requests and token usage, and ATIF must contain the
+expected agent response.

adapters/codex/README.md

@@ -3,7 +3,7 @@ SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All
 SPDX-License-Identifier: Apache-2.0
 -->
 
-# Codex Adapter
+# NVIDIA NeMo Fabric Codex Adapter
 
 The `nvidia.fabric.codex` adapter uses the official Codex Python SDK behind
 Fabric's normalized invocation contract. It does not resolve or execute a
@@ -12,11 +12,16 @@ app-server runtime and typed JSON-RPC protocol.
 
 ## Install
 
-Build the local wheels and install the Codex adapter:
+To install just the Codex adapter by itself:
 
 ```bash
-just wheels
-python -m pip install --find-links dist "nemo-fabric[codex]"
+pip install "nemo-fabric[codex]"
+```
+
+To install just the Codex adapter along with the NeMo Fabric Runtime:
+
+```bash
+pip install "nemo-fabric[codex, runtime]"
 ```
 
 ## Authentication
@@ -134,7 +139,7 @@ cargo install --locked --path nemo-relay/crates/cli
 ```
 
 Removal of this temporary source installation is tracked in
-[TODO.md](../../TODO.md#nemo-relay-06x-request-decoding-release).
+[TODO.md](https://github.com/NVIDIA/NeMo-Fabric/blob/main/TODO.md#nemo-relay-06x-request-decoding-release).
 
 The `nemo-relay` Python package does not install this executable. Refer to the
 [NeMo Relay installation guide](https://docs.nvidia.com/nemo/relay/getting-started/installation)
@@ -153,24 +158,3 @@ Relay OpenInference provides the semantic chain, LLM, and tool hierarchy with
 decoded prompt, response, and token attributes. Prefer Relay OpenInference for
 agent-turn inspection.
 
-## Local Validation
-
-Run the unit and opt-in real SDK tests separately:
-
-```bash
-uv run pytest tests/adapters/test_codex_adapter.py -q
-RUN_FABRIC_CODEX_INTEGRATION=1 uv run pytest tests/e2e/test_codex.py -q
-RUN_FABRIC_CODEX_RELAY_INTEGRATION=1 \
-  FABRIC_TEST_NEMO_RELAY_COMMAND=/path/to/nemo-relay \
-  uv run pytest tests/e2e/test_codex.py -q
-```
-
-Set `FABRIC_TEST_CODEX_BIN=/path/to/codex` on either opt-in command to validate
-an explicit app-server override instead of the SDK-pinned runtime.
-
-The SDK test uses the current Codex authentication state and exercises both a
-one-shot invocation and multi-turn thread resume. The Relay test additionally
-requires an external gateway binary and verifies one-shot and resumed model
-responses, stable thread identity, ATOF, and ATIF; gateway startup alone is not
-a passing result. The semantic regression also requires decoded LLM request
-content, a model, token usage, and the expected agent response in ATIF.

adapters/codex/testing.md

@@ -0,0 +1,26 @@
+<!--
+SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+SPDX-License-Identifier: Apache-2.0
+-->
+
+# Testing the Codex Adapter
+
+Run the unit and opt-in real SDK tests separately:
+
+```bash
+uv run pytest tests/adapters/test_codex_adapter.py -q
+RUN_FABRIC_CODEX_INTEGRATION=1 uv run pytest tests/e2e/test_codex.py -q
+RUN_FABRIC_CODEX_RELAY_INTEGRATION=1 \
+  FABRIC_TEST_NEMO_RELAY_COMMAND=/path/to/nemo-relay \
+  uv run pytest tests/e2e/test_codex.py -q
+```
+
+Set `FABRIC_TEST_CODEX_BIN=/path/to/codex` on either opt-in command to validate
+an explicit app-server override instead of the SDK-pinned runtime.
+
+The SDK test uses the current Codex authentication state and exercises both a
+one-shot invocation and multi-turn thread resume. The Relay test additionally
+requires an external gateway binary and verifies one-shot and resumed model
+responses, stable thread identity, ATOF, and ATIF; gateway startup alone is not
+a passing result. The semantic regression also requires decoded LLM request
+content, a model, token usage, and the expected agent response in ATIF.

adapters/common/README.md

@@ -15,7 +15,12 @@ Install the package directly when developing an adapter:
 pip install nemo-fabric-adapters-common
 ```
 
+Alternately through the NeMo Fabric metapackage:
+
+```bash
+pip install "nemo-fabric[adapters-common]"
+```
+
 Refer to the [NeMo Fabric documentation](https://nvidia-nemo-fabric.docs.buildwithfern.com/nemo/fabric)
 for adapter and configuration guidance. Source code is available in the
 [NVIDIA NeMo Fabric repository](https://github.com/NVIDIA/nemo-fabric/).
-

adapters/deepagents/README.md

@@ -3,16 +3,22 @@ SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All
 SPDX-License-Identifier: Apache-2.0
 -->
 
-# LangChain Deep Agents Adapter
+# NVIDIA NeMo Fabric LangChain Deep Agents Adapter
 
 Runs a [LangChain Deep Agents](https://github.com/langchain-ai/deepagents) agent
 through Fabric's inline Python adapter lifecycle. The same adapter supports
 one-shot, multi-turn, and resumed execution.
 
-Install Fabric with the adapter dependency before running it:
+To install just the Deep Agents adapter by itself:
 
 ```bash
-python3 -m pip install -e ".[deepagents]"
+pip install "nemo-fabric[deepagents]"
+```
+
+To install just the Deep Agents adapter along with the NeMo Fabric Runtime:
+
+```bash
+pip install "nemo-fabric[deepagents, runtime]"
 ```
 
 ## Model and Authentication

adapters/hermes/README.md

@@ -3,18 +3,27 @@ SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All
 SPDX-License-Identifier: Apache-2.0
 -->
 
-# Hermes Agent Adapter
+# NVIDIA NeMo Fabric Hermes Agent Adapter
 
 This adapter runs Hermes Agent through its Python SDK.
 
-Fabric invokes the adapter module with `python -m` through the core runtime
-lifecycle. The module entry point and the descriptor's callable route use the
-same `run(payload: dict) -> dict` implementation.
+## Install
+
+To install just the Hermes adapter by itself:
+
+```bash
+pip install "nemo-fabric[adapters-hermes]"
+```
+
+To install just the Hermes adapter along with the NeMo Fabric Runtime:
+
+```bash
+pip install "nemo-fabric[adapters-hermes, runtime]"
+```
 
 ## What It Maps
 
-The adapter receives Fabric's normalized payload and materializes Hermes-native
-configuration for:
+The adapter receives a normalized payload from Fabric and materializes Hermes-native configuration for:
 
 - model provider, model name, base URL, temperature, and token settings;
 - workspace and terminal settings;

pypi.md

@@ -0,0 +1,76 @@
+<!--
+SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+SPDX-License-Identifier: Apache-2.0
+-->
+
+# NVIDIA NeMo Fabric
+
+NVIDIA NeMo Fabric is a runtime execution layer for agents. It turns multiple
+agent harnesses into one configurable, observable lifecycle surface.
+
+NeMo Fabric standardizes how applications configure, launch, invoke, and
+collect artifacts from agent harnesses. It provides:
+
+- a versioned, typed configuration contract;
+- profile-based configuration for evaluation and ablation runs;
+- adapter integrations for harness-specific launch and control;
+- a Python SDK backed by a Rust core;
+- normalized run results, artifact manifests, and telemetry references.
+
+## Install
+
+Install the core runtime and Python SDK:
+
+```bash
+pip install "nemo-fabric[runtime]"
+```
+
+To use a supported agent harness, install its adapter extra:
+
+```bash
+pip install "nemo-fabric[claude]"
+pip install "nemo-fabric[codex]"
+pip install "nemo-fabric[deepagents]"
+pip install "nemo-fabric[hermes]"
+```
+
+Fabric supports running an agent harness in a different virtual environment than the one used to run Fabric itself. This is useful for running agents that have conflicting dependencies with Fabric or other agents.
+
+The adapter must be installed into the virtual environment that the harness is installed in. For this reason adapters intentionally have minimal dependencies.
+
+### Integrations
+
+Harbor integration:
+
+```bash
+pip install "nemo-fabric[harbor]"
+```
+
+Relay integration:
+
+```bash
+pip install "nemo-fabric[relay]"
+```
+
+### Python Versions
+
+NeMo Fabric supports Python versions 3.11-3.14, however some of the integrations and adapters may have additional requirements. Specifically the Hermes adapter doesn't support Python 3.14 yet, and the Harbor integration requires Python 3.12 or later.
+
+## Core Concepts
+
+- **Agent source:** Provide either an agent package path or a typed
+  `FabricConfig`.
+- **Typed configuration:** Construct configuration in memory with the Python
+  SDK, or use `agent.yaml` as a portable representation.
+- **Profiles:** Vary the harness, model, MCP servers, tools, skills, telemetry,
+  or environment without editing the base configuration.
+- **Adapters:** Select harness-specific integrations with
+  `harness.adapter_id`.
+- **Artifacts:** Receive normalized output, logs, patches, and telemetry
+  references through an `ArtifactManifest`.
+
+## Learn More
+
+Refer to the [NVIDIA NeMo Fabric documentation](https://nvidia-nemo-fabric.docs.buildwithfern.com/nemo/fabric)
+for installation, configuration, and usage guidance. Source code is available
+in the [NVIDIA NeMo Fabric repository](https://github.com/NVIDIA/nemo-fabric/).

pyproject.toml

@@ -22,7 +22,7 @@ classifiers = [
 ]
 license = "Apache-2.0"
 license-files = ["LICENSE"]
-readme = "README.md"
+readme = "pypi.md"
 requires-python = ">=3.11"
 dependencies = []