Instructions to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Use Docker

docker model run hf.co/aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

LM Studio
Jan

vLLM

How to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "aogavrilov/diffusiongemma-agent-iq3-cuda13"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "aogavrilov/diffusiongemma-agent-iq3-cuda13",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Ollama
How to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with Ollama:
```
ollama run hf.co/aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M
```

Unsloth Studio

How to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for aogavrilov/diffusiongemma-agent-iq3-cuda13 to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for aogavrilov/diffusiongemma-agent-iq3-cuda13 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for aogavrilov/diffusiongemma-agent-iq3-cuda13 to start chatting

How to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with Docker Model Runner:
```
docker model run hf.co/aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M
```

Lemonade

How to use aogavrilov/diffusiongemma-agent-iq3-cuda13 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull aogavrilov/diffusiongemma-agent-iq3-cuda13:Q4_K_M

Run and chat with the model

lemonade run user.diffusiongemma-agent-iq3-cuda13-Q4_K_M

List all available models

lemonade list

diffusiongemma-agent-iq3-cuda13 / payload /app /docs /ide_client_profiles.md

aogavrilov

Add files using upload-large-folder tool

ef2127a verified 16 days ago

preview code

Raw

History Blame Contribute Delete

19.1 kB

IDE Client Profiles

The local model is exposed through LiteLLM as a normal OpenAI-compatible endpoint:

base_url: http://127.0.0.1:4100/v1
api_key: dummy
model: diffusiongemma-local

Use these profile files:

configs/client_profiles/openai-compatible.local.json
configs/client_profiles/openai.env
configs/client_profiles/aider.dg-workspace.conf.yml
configs/client_profiles/continue.config.yaml
configs/client_profiles/agent-client-pack.json
configs/client_profiles/openhands.dg.toml
configs/client_profiles/openhands.env
configs/client_profiles/swe-agent.dg.yaml
configs/client_profiles/mini-swe-agent.dg.yaml
configs/client_profiles/mcp-server.json
configs/client_profiles/mcp-client-snippets.json
configs/client_profiles/claude-code.mcp.json
configs/client_profiles/claude-desktop-mcp.json
configs/client_profiles/cursor.mcp.json
configs/client_profiles/vscode.mcp.json
configs/client_profiles/agent-instructions.md
configs/client_profiles/AGENTS.dg.md
configs/client_profiles/CLAUDE.dg.md
configs/client_profiles/copilot-instructions.dg.md
configs/client_profiles/diffusiongemma.instructions.md
configs/client_profiles/cursor-rules.dg.mdc
configs/client_profiles/litellm-local-model-registry.json

Generate or inspect the full client pack:

scripts/dg_agent.sh client-pack
scripts/dg_agent.sh client-pack --json
scripts/dg_agent.sh client-pack --env
scripts/dg_agent.sh client-pack --write

The client pack includes settings for OpenAI SDK, Aider, OpenCode, Goose, Continue, Cline, Roo Code, Kilo Code, MCP-capable IDE clients, and optional OpenHands/SWE-agent routes through the local LiteLLM endpoint.

One-Shot Client Init

For Cursor, Claude Code, Claude Desktop, or VS Code, use the high-level bootstrap:

scripts/dg_agent.sh client-init --repo /path/to/repo --client cursor
scripts/dg_agent.sh client-smoke --repo /path/to/repo --client cursor --live
scripts/dg_agent.sh client-report --repo /path/to/repo --client cursor --live
scripts/dg_agent.sh agent-commands --repo /path/to/repo --target all

By default this writes .dg-agent/, adds the DG MCP server and Repomix, then adds Serena only when serena-mcp --check-installed succeeds. It also writes AGENTS, Claude, Copilot, VS Code, and Cursor instruction files. The current WSL Serena runtime passes that check, so the active default bundle is diffusiongemma-local-agent + repomix + serena.

Run client-smoke after bootstrap when you want a readiness gate before attaching an external IDE or ACP client. It verifies the repo-local hub, MCP client config, agent rules, launchers, and with --live also checks the backend, proxy, and LiteLLM endpoints.

Run client-report when you want a portable handoff for another agent or IDE session. It writes .dg-agent/CLIENT_HANDOFF.md and .dg-agent/client-handoff.json with ready commands, MCP config, route map, latest capability snapshot, and optional live endpoint status.

Run agent-commands to install the repo command layer. It keeps generic workflow snippets in .dg-agent/commands/ and writes the Claude Code project skill .claude/skills/dg-local-agent/SKILL.md.

For ACP-capable clients, use the bridge command instead of wiring the OSS agent server by hand:

scripts/dg_agent.sh agent-bridge --repo /path/to/repo --server opencode-acp
scripts/dg_agent.sh agent-bridge --repo /path/to/repo --server opencode-acp --start
scripts/dg_agent.sh agent-bridge --repo /path/to/repo --server openhands-acp

The default bridge prepares the repo through client-init, then exposes it via the upstream OpenCode ACP server with DG MCP and Repomix mounted. Add the separate Serena MCP entry in an IDE when semantic/LSP tools are needed. --server goose-serve, --server goose-acp, and --server openhands-acp use Goose or OpenHands ACP modes instead.

After any workspace bootstrap, open .dg-agent/AGENT_HUB.md or run .dg-agent/bin/hub. That handoff is the shortest route map for humans and external agents; .dg-agent/bin/hub --json prints the same data as JSON.

Repo-Local Workspace Init

For a target repository, write a local .dg-agent/ directory:

scripts/dg_agent.sh workspace-init --repo /path/to/repo

It creates:

.dg-agent/client-pack.json
.dg-agent/AGENT_HUB.md
.dg-agent/agent-hub.json
.dg-agent/COMMANDS.md
.dg-agent/command-kit.json
.dg-agent/commands/dg-report.md
.dg-agent/commands/dg-smoke.md
.dg-agent/commands/dg-context.md
.dg-agent/commands/dg-plan-task.md
.dg-agent/commands/dg-agent.md
.dg-agent/commands/dg-verify.md
.dg-agent/commands/dg-mcp-handoff.md
.dg-agent/commands/dg-codex.md
.dg-agent/claude-skill/SKILL.md
.dg-agent/CODEX.md
.dg-agent/codex.config.toml
.dg-agent/codex.env
.dg-agent/IDE_CLIENTS.md
.dg-agent/ide-client-snippets.json
.dg-agent/openai-compatible.local.json
.dg-agent/openai.env
.dg-agent/kilo-code.config.json
.dg-agent/env.sh
.dg-agent/README.md
.dg-agent/aider.dg-fast.conf.yml
.dg-agent/aider.dg-model-settings.yml
.dg-agent/aider.dg-model-metadata.json
.dg-agent/continue.config.yaml
.dg-agent/opencode.dg.json
.dg-agent/opencode.dg-mcp.json
.dg-agent/openhands.dg.toml
.dg-agent/openhands.env
.dg-agent/swe-agent.dg.yaml
.dg-agent/mini-swe-agent.dg.yaml
.dg-agent/mcp-server.json
.dg-agent/mcp-client-snippets.json
.dg-agent/claude-code.mcp.json
.dg-agent/claude-desktop-mcp.json
.dg-agent/cursor.mcp.json
.dg-agent/vscode.mcp.json
.dg-agent/agent-instructions.md
.dg-agent/AGENTS.dg.md
.dg-agent/CLAUDE.dg.md
.dg-agent/copilot-instructions.dg.md
.dg-agent/diffusiongemma.instructions.md
.dg-agent/cursor-rules.dg.mdc
.dg-agent/goose-mcp.dg.yaml
.dg-agent/litellm-local-model-registry.json
.dg-agent/bin/agent
.dg-agent/bin/client-smoke
.dg-agent/bin/client-report
.dg-agent/bin/context
.dg-agent/bin/verify
.dg-agent/bin/status
.dg-agent/bin/aider
.dg-agent/bin/opencode
.dg-agent/bin/opencode-mcp
.dg-agent/bin/opencode-acp
.dg-agent/bin/goose
.dg-agent/bin/goose-mcp
.dg-agent/bin/mcp
.dg-agent/bin/serena-mcp
.dg-agent/bin/client-init
.dg-agent/bin/agent-bridge
.dg-agent/bin/hub
.dg-agent/bin/agent-commands

For git repositories, workspace-init writes both .dg-agent/ and .serena/ to the repo-local .git/info/exclude. Serena may create .serena/project.yml on first semantic-MCP startup, and this keeps that local metadata out of normal git status.

Load the env from the target repo:

set -a
. .dg-agent/env.sh
set +a

Then use the repo-local launchers:

.dg-agent/bin/status
.dg-agent/bin/agent --task "..." --file path/to/file
.dg-agent/bin/aider app.py --message "Make the smallest patch"
.dg-agent/bin/context --task "..." --max-files 3
.dg-agent/bin/verify --file path/to/file
.dg-agent/bin/openhands --task "..." --dry-run
.dg-agent/bin/openhands-acp --dry-run
.dg-agent/bin/swe-agent --task "..." --dry-run
.dg-agent/bin/mini-swe-agent --task "..." --dry-run
.dg-agent/bin/mini-swe-run --task "..." --dry-run --json
.dg-agent/bin/mini-swe-runs list
.dg-agent/bin/mcp --list-tools
.dg-agent/bin/mcp-http --host 127.0.0.1 --port 8765
.dg-agent/bin/serena-mcp --help-local
.dg-agent/bin/client-init --client cursor
.dg-agent/bin/client-smoke --client cursor --live
.dg-agent/bin/client-report --client cursor --live
.dg-agent/bin/agent-commands --target all
.dg-agent/bin/agent-bridge --server opencode-acp
.dg-agent/bin/agent-bridge --server openhands-acp
.dg-agent/bin/hub
.dg-agent/bin/goose-mcp --help-local
.dg-agent/bin/agent-rules --target all

Check readiness before a live edit:

scripts/dg_agent.sh preflight --repo /path/to/repo --task "..." --file path

Use --force only when you intentionally want to overwrite changed .dg-agent/ files.

OpenAI SDK

Bash:

set -a
. /root/diffusiongemma-agent/configs/client_profiles/openai.env
set +a

Python client shape:

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:4100/v1",
    api_key="dummy",
)

models = client.models.list()

Continue

Use:

configs/client_profiles/continue.config.yaml

The profile uses Continue's openai provider pointed at the local LiteLLM gateway.

Codex CLI

Use the project-local Codex profile when Codex CLI should talk to the local DG safe agent proxy:

.dg-agent/bin/codex-profile --target all
source .dg-agent/codex.env
codex

This writes .codex/config.toml from .dg-agent/codex.config.toml. The default provider is the safe proxy at http://127.0.0.1:8090/v1; for plain chat, switch model_provider to diffusiongemma-local-chat in the generated config.

Aider

Preferred reliable launcher:

.dg-agent/bin/aider app.py --message "Make the smallest patch"

Direct upstream Aider config:

aider --config .dg-agent/aider.dg-fast.conf.yml app.py

The config uses the DG Aider proxy at http://127.0.0.1:8090/v1, disables repo map expansion, keeps chat history small, disables auto commits/lint/test, and uses .dg-agent/aider.dg-model-settings.yml plus .dg-agent/aider.dg-model-metadata.json for local zero-cost model metadata.

Cline / Roo Code / Kilo Code

Open .dg-agent/IDE_CLIENTS.md or .dg-agent/ide-client-snippets.json after workspace-init. They include two routes: the LiteLLM chat/edit endpoint and the safe agent proxy endpoint for command-like tool delegation.

Configure the provider as OpenAI-compatible for the safe agent route:

Provider: OpenAI Compatible
Base URL: http://127.0.0.1:8090/v1
API key: dummy
Model: diffusiongemma-26b-a4b-it-iq4xs-aider-local

For simple chat/edit flows that do not need command delegation, use:

Base URL: http://127.0.0.1:4100/v1
Model: diffusiongemma-local

For Kilo Code, use .dg-agent/kilo-code.config.json as the local OpenAI-compatible custom provider template. Keep requests small. The current runtime is tuned for file-level tasks and short context packs, not repository-wide chat history.

MCP-capable clients

Use the local MCP SDK server when a client can mount stdio MCP tools. This is the strongest wrapper path for existing clients because the model does not need to invent tool JSON; clients call stable tools that delegate to the repository agent pipeline.

When a client supports MCP over HTTP but cannot spawn local stdio processes, run the streamable HTTP launcher instead:

.dg-agent/bin/mcp-http --host 127.0.0.1 --port 8765

Endpoint:

http://127.0.0.1:8765/mcp

Server command:

/root/diffusiongemma-agent/scripts/run_mcp_server.sh

Copy-ready templates:

Claude Code project .mcp.json: .dg-agent/claude-code.mcp.json
Claude Desktop claude_desktop_config.json: .dg-agent/claude-desktop-mcp.json
Cursor project .cursor/mcp.json: .dg-agent/cursor.mcp.json
VS Code workspace .vscode/mcp.json: .dg-agent/vscode.mcp.json
All snippets: .dg-agent/mcp-client-snippets.json

Install or merge the server entry from a target repo:

.dg-agent/bin/mcp-client-config --client claude-code
.dg-agent/bin/mcp-client-config --client cursor
.dg-agent/bin/mcp-client-config --client cursor --with-repomix
.dg-agent/bin/mcp-client-config --client cursor --with-serena
.dg-agent/bin/mcp-client-config --client cursor --with-repomix --with-serena
.dg-agent/bin/mcp-client-config --client cursor --with-oss-stack
.dg-agent/bin/mcp-client-config --client vscode

The merge keeps existing unrelated MCP servers. If the diffusiongemma-local-agent entry already exists with different settings, use --force only when you intentionally want to replace it.

The exported MCP tools are dg_repo_status, dg_list_files, dg_search, dg_read_file, dg_git_diff, dg_task_note, dg_task_notes, dg_status, dg_context, dg_rag_context, dg_rag_answer, dg_repo_pack, dg_repo_map, dg_ast_grep, dg_code_outline, dg_preflight, dg_plan, dg_task, dg_session, dg_verify, dg_capabilities, dg_client_smoke, dg_client_report, dg_sessions, and dg_session_artifact. Use the repo tools for navigation and inspection, use dg_rag_context for compact repo-scale retrieval, use dg_repo_pack for filtered Repomix packed context, save handoff notes with dg_task_note, generate client handoff with dg_client_report, then run either the one-shot dg_session path or the stepwise dg_plan -> dg_task path.

mcp-client-snippets.json also carries optional native Repomix MCP server and Serena MCP server entries. Copy Repomix when the IDE should talk to upstream repository packing directly. Copy Serena when the IDE should use semantic/LSP tools such as symbol overview, references, diagnostics, renames, and safe symbol edits in addition to the DG MCP tool bridge.

Agent Rules

Install repo-local instruction files for clients that read project rules:

.dg-agent/bin/agent-rules --target all

Targets:

AGENTS.md
CLAUDE.md
.github/copilot-instructions.md
.github/instructions/diffusiongemma.instructions.md
.cursor/rules/diffusiongemma-local-agent.mdc

Existing AGENTS.md, CLAUDE.md, and Copilot instructions are preserved with a marked DG block. Dedicated Cursor and VS Code generated files are not replaced unless --force is used.

OpenHands

Use the LiteLLM Proxy profile:

Config: configs/client_profiles/openhands.dg.toml
Model: litellm_proxy/diffusiongemma-local
Base URL: http://127.0.0.1:4100
API key: dummy

For a target repo initialized with workspace-init, use the copied files:

.dg-agent/openhands.dg.toml
.dg-agent/openhands.env

Launcher:

.dg-agent/bin/openhands --task "..." --dry-run
.dg-agent/bin/openhands-acp --dry-run
.dg-agent/bin/openhands-mcp --reset
.dg-agent/bin/qwen-code --dry-run
.dg-agent/bin/autogen --dry-run
.dg-agent/bin/smolagents --dry-run
.dg-agent/bin/langgraph --dry-run
.dg-agent/bin/crewai --dry-run
.dg-agent/bin/open-interpreter --dry-run
.dg-agent/bin/llamaindex --dry-run
.dg-agent/bin/haystack --dry-run

For ACP clients, use .dg-agent/bin/openhands-acp or .dg-agent/bin/agent-bridge --server openhands-acp. This route uses openhands acp --override-with-envs; avoid the standalone openhands-acp entrypoint from the package.

For OpenHands' built-in MCP management, run .dg-agent/bin/openhands-mcp --reset. It writes .dg-agent/openhands-persistence/mcp.json with the DG MCP server, Repomix, and Serena bound to the current repo.

For Qwen Code, run .dg-agent/bin/qwen-code --dry-run first. On this Windows host the launcher prefers the WSL MCP route with DG, Repomix, and Serena; native PowerShell falls back to explicit read-only mode on the safe GPU gateway. Name a file in the prompt and use Aider/session for edits.

For AutoGen AgentChat, run .dg-agent/bin/autogen --dry-run or .dg-agent/bin/autogen --smoke-import. The launcher uses .dg-agent/autogen.dg.json with OpenAIChatCompletionClient.

For Hugging Face smolagents, run .dg-agent/bin/smolagents --dry-run or .dg-agent/bin/smolagents --smoke-import. The launcher uses .dg-agent/smolagents.dg.json with CodeAgent and OpenAIModel.

For LangGraph/LangChain, run .dg-agent/bin/langgraph --dry-run or .dg-agent/bin/langgraph --smoke-import. The launcher uses .dg-agent/langgraph.dg.json with ChatOpenAI and a LangGraph/LangChain agent factory.

For CrewAI, run .dg-agent/bin/crewai --dry-run or .dg-agent/bin/crewai --smoke-import. The launcher uses .dg-agent/crewai.dg.json with CrewAI Agent, Task, Crew, and LLM.

For Open Interpreter, run .dg-agent/bin/open-interpreter --dry-run or .dg-agent/bin/open-interpreter --smoke-import. The launcher uses .dg-agent/open-interpreter.dg.json with auto_run=false and safe_mode=ask.

For LlamaIndex, run .dg-agent/bin/llamaindex --dry-run or .dg-agent/bin/llamaindex --smoke-import. The launcher uses .dg-agent/llamaindex.dg.json with OpenAILike, AgentWorkflow.from_tools_or_functions, ReActAgent, and bounded repo tools.

For Haystack, run .dg-agent/bin/haystack --dry-run or .dg-agent/bin/haystack --smoke-import. The launcher uses .dg-agent/haystack.dg.json with InMemoryDocumentStore, InMemoryBM25Retriever, and OpenAIChatGenerator.

Install the upstream CLI only when you want to run the real external loop:

scripts/dg_agent.sh bootstrap --only openhands --install

SWE-agent / mini-swe-agent

Use the OpenAI-compatible LiteLLM endpoint:

Config: configs/client_profiles/swe-agent.dg.yaml
Config: configs/client_profiles/mini-swe-agent.dg.yaml
Model: openai/diffusiongemma-local
Base URL: http://127.0.0.1:4100/v1
API key: dummy

The mini-swe-agent profile keeps the loop short and sets cost tracking to ignore_errors, which is more practical for a local zero-billing model. These profiles are experimental; the default reliable path remains scripts/dg_agent.sh agent/session/task.

Launchers:

.dg-agent/bin/swe-agent --task "..." --dry-run
.dg-agent/bin/mini-swe-agent --task "..." --dry-run
.dg-agent/bin/mini-swe-run --task "..." --dry-run --json
.dg-agent/bin/mini-swe-runs show --latest

Prefer mini-swe-run when you want preserved artifacts. It writes a report, logs, command file, and trajectory path under runlogs/mini-swe-agent/. Use mini-swe-runs to list, show, or print those artifacts.

Install selected upstream CLIs:

scripts/dg_agent.sh bootstrap --only mini-swe-agent --install
scripts/dg_agent.sh bootstrap --only swe-agent --install

Smoke

scripts/dg_agent.sh smoke --suite gateway-clients
scripts/dg_agent.sh smoke --suite openai-sdk
scripts/dg_agent.sh smoke --suite client-pack
scripts/dg_agent.sh smoke --suite workspace-init
scripts/dg_agent.sh smoke --suite client-init
scripts/dg_agent.sh smoke --suite client-smoke
scripts/dg_agent.sh smoke --suite client-report
scripts/dg_agent.sh smoke --suite agent-commands
scripts/dg_agent.sh smoke --suite codex-profile
scripts/dg_agent.sh smoke --suite ide-clients
scripts/dg_agent.sh smoke --suite agent-bridge
scripts/dg_agent.sh smoke --suite openhands-acp
scripts/dg_agent.sh smoke --suite openhands-mcp
scripts/dg_agent.sh smoke --suite qwen-code
scripts/dg_agent.sh smoke --suite autogen
scripts/dg_agent.sh smoke --suite smolagents
scripts/dg_agent.sh smoke --suite langgraph
scripts/dg_agent.sh smoke --suite crewai
scripts/dg_agent.sh smoke --suite open-interpreter
scripts/dg_agent.sh smoke --suite llamaindex
scripts/dg_agent.sh smoke --suite haystack
scripts/dg_agent.sh smoke --suite preflight
scripts/dg_agent.sh smoke --suite external-agents

The smoke checks the live LiteLLM model registry and verifies that the checked in client profiles match the actual endpoint.

The openai-sdk smoke uses the installed Python OpenAI SDK against http://127.0.0.1:4100/v1, sends one bounded chat completion, and verifies that backend, proxy, and LiteLLM remain healthy afterwards.

The gateway intentionally runs generic chat in safe compatibility mode by default. For real code edits, prefer scripts/dg_agent.sh session or scripts/dg_agent.sh task; raw IDE agent loops are still experimental with the current model profile.