Draft: Integrate liked prs

CONTRIBUTING.md

@@ -50,7 +50,8 @@ brew install go node uv
 3. Add or update tests for changed behavior.
 4. Update the documentation when you change commands, flags, output, or
    requirements.
-5. Run `make test`.
+5. Run `make test`. It runs the engine tests in parallel; set
+   `PYTEST_XDIST_AUTO_NUM_WORKERS` to choose the worker count.
 
 ## Open a pull request
 

Makefile

@@ -3,10 +3,10 @@
 # The OpenCode client is built for Apple Silicon only.
 ifeq ($(shell uname -sm),Darwin arm64)
 test: test-cli test-engine
-CLIENTS = opencode pi
+CLIENTS = opencode pi omp
 else
 test: test-cli
-CLIENTS = pi
+CLIENTS = pi omp
 endif
 
 test-cli:
@@ -24,7 +24,7 @@ PYTHON_ROOT = $(shell uv python dir)/cpython-3.12.12-macos-aarch64-none
 
 test-engine:
 	cd engine && uv venv --allow-existing --python 3.12 && uv pip install '.[validation]'
-	cd engine && .venv/bin/pytest -v -rs
+	cd engine && nice -n 10 .venv/bin/pytest -v -rs -n auto --dist loadgroup
 	uv python install 3.12.12
 	rm -rf .build/bundle-mlx .build/smoke
 	$(PYTHON_ROOT)/bin/python3.12 packaging/build-mlx-runtime.py --python-root $(PYTHON_ROOT) --output .build/bundle-mlx

README.md

@@ -48,11 +48,12 @@ terminals. All other data lives under `~/.local/share/rec`.
 
 We find the best performance from our models when paired with the
 [pi](https://pi.dev/) harness. For convenience, rec provides pre-configured
-recipes for running pi and [OpenCode](https://opencode.ai) harnesses. Run
+recipes for running pi, omp, and [OpenCode](https://opencode.ai) harnesses. Run
 them using:
 
 ```bash
 rec launch pi  # preferred
+rec launch omp
 rec launch opencode
 ```
 
@@ -65,7 +66,7 @@ run downloads the model and harness. The model server continues to run after
 you close the terminal. Run `rec stop` to stop it.
 
 These harnesses are kept separate from anything on your system. Use
-`rec exec pi|opencode ARGS` to run their own commands, for example:
+`rec exec pi|omp|opencode ARGS` to run their own commands, for example:
 
 ```bash
 rec exec pi install npm:PACKAGE
@@ -109,6 +110,18 @@ The response contains generated text in `choices`. For other clients, use
 base URL `http://127.0.0.1:11500/v1`, model `flint`, and the API key from
 `rec auth token`.
 
+To bound the reasoning of a chat request, add `"thinking_budget": N`, a
+nonnegative integer. It needs `grammar=flint` (the default). N counts the
+tokens after the model's own `<think>`; when N is reached, Flint closes the
+`</think>` block itself. A budget of 0 skips reasoning. A prompt that already
+ends with `<think>` is rejected with HTTP 400. A prompt with an
+unclosed `<think>` anywhere is also rejected. `/v1/messages` rejects only
+`thinking_budget` with 400; `thinking.budget_tokens` is ignored as before. When
+the request sets `thinking_budget` (0 included), the response `usage` reports
+`reasoning_tokens` and `thinking_forced_close`. Streaming responses report them
+only in the final usage chunk, so set `stream_options.include_usage`. Without a
+budget these keys do not appear.
+
 The Anthropic-compatible API is available at
 `http://127.0.0.1:11500/v1/messages`. Point `ANTHROPIC_BASE_URL` at
 `http://127.0.0.1:11500` and send the token from `rec auth token` as a bearer
@@ -119,8 +132,8 @@ token or as `x-api-key`. The model is `flint`.
 
 | Command | What it does |
 |---|---|
-| `rec launch pi\|opencode` | Start the server if needed and open a harness |
-| `rec exec pi\|opencode ARGS` | Run a harness command, such as installing a plugin |
+| `rec launch pi\|omp\|opencode` | Start the server if needed and open a harness |
+| `rec exec pi\|omp\|opencode ARGS` | Run a harness command, such as installing a plugin |
 | `rec serve` | Start the model server in this terminal |
 | `rec status` | Show whether the server is running |
 | `rec stop` | Stop the server |

cli/clients.go

@@ -14,6 +14,12 @@ var openCodeLogo embed.FS
 //go:embed clients/pi/package.json
 var piPackage []byte
 
+//go:embed clients/omp/package.json
+var ompPackage []byte
+
+//go:embed clients/omp/package-lock.json
+var ompLock []byte
+
 //go:embed clients/pi/package-lock.json
 var piLock []byte