env2llm 0.1.8

Generate environment maps (services, artifacts, env vars) for LLM decision-making

env2llm

AI Cost Tracking

• 🤖 LLM usage: $1.1624 (4 commits)
• 👤 Human dev: ~$287 (2.9h @ $100/h, 30min dedup)

Generated on 2026-06-06 using openrouter/qwen/qwen3-coder-next

---

Generate environment maps for LLM agents: available services, commands, artifacts, masked environment variables, and runtime policies.

Default output is DOQL-flavored LESS (environment.doql.less), compatible with nlp2dsl examples:

environment[name="03-report-and-notify"] {
  NLP2DSL_BACKEND_URL: "http://localhost:8010";
  LLM_MODEL: "openrouter/...";
}

runtimes[0] { id: "orchestrator:nlp-service"; kind: "orchestrator"; ... }
commands[3] { name: "generate_report"; required: "report_type"; ... }

Install

cd env2llm
pip install -e ".[dev]"

CLI

# From an nlp2dsl example directory
env2llm /path/to/nlp2dsl/examples/03-report-and-notify

# Other formats
env2llm . --format yaml
env2llm . --format json
env2llm . --format markdown

# Optional: attach live Linux desktop snapshot (GNOME/X11 via wmctrl/xdotool)
env2llm . --probe-desktop
# or: ENV2LLM_DESKTOP_PROBE=1 env2llm .

# Koru projects: auto-attach MCP tool catalog (koru_list_tickets, …) when koruapi installed
env2llm . --probe-mcp
# or auto when cwd is a Koru repo with koruapi on PYTHONPATH

Writes to .nlp2dsl/registry/environment.<ext> (and mirrors legacy paths for DOQL).

Desktop / GUI probe (optional)

When --probe-desktop or ENV2LLM_DESKTOP_PROBE=1 is set, env2llm adds:

• desktop { ... } and desktop_windows[N] { title, x, y, width, height, ... } DOQL blocks
• runtime probe:desktop (desktop://session) for GUI automation via nlp2uri
• commands desktop_focus_window, desktop_move_window, desktop_screenshot_*, desktop_open_app
• summary keys in data (desktop.window_titles, desktop.browser_windows, …)

Requires (Linux): wmctrl and/or xdotool on a graphical session. Headless hosts get status: unknown without failing workflow maps.

Browser window titles are detected heuristically (Firefox, Chrome, Edge, …). Page DOM/content is not scraped — use testql / Playwright for that layer.

Python API

from env2llm import RegistryService, ensure_environment_map, render_format, generate_system_map

path = ensure_environment_map("examples/03-report-and-notify")
ir = generate_system_map("examples/03-report-and-notify", example_id="03-report-and-notify")
yaml_text = render_format(ir, "yaml")

# Live registry service (load / refresh / render / MQTT)
service = RegistryService(".", project_id="my-app")
registry_json = service.render("json")

REST API (env2llm-serve)

env2llm-serve --project . --port 8770
# optional MQTT fan-out:
ENV2LLM_MQTT_ENABLED=1 env2llm-serve --project . --mqtt

Method	Path	Opis
GET	/health	status
GET	/v1/registry?format=json|yaml|doql|md&refresh=1	registry
POST	/v1/registry/refresh	regenerate + persist (+ MQTT)
GET	/v1/registry/desktop	desktop probe slice
GET	/v1/registry/commands	command schemas
GET	/v1/registry/uris	nlp2uri URI index (needs nlp2uri[envmap])
GET	/v1/registry/mqtt	MQTT bridge status

MCP (env2llm-mcp)

{
  "mcpServers": {
    "env2llm": {
      "command": "env2llm-mcp",
      "args": ["--project", "/path/to/project"],
      "env": { "ENV2LLM_MQTT_ENABLED": "1" }
    }
  }
}

Tools: env2llm_get_registry, env2llm_render_registry, env2llm_refresh_registry, env2llm_get_desktop, env2llm_list_commands, env2llm_list_uris, env2llm_mqtt_status.

MQTT (env2llm-mqtt)

Requires pip install 'env2llm[mqtt]' (paho-mqtt).

# Standalone bridge: listen for refresh, publish retained snapshots
ENV2LLM_MQTT_ENABLED=1 env2llm-mqtt bridge --project .

# One-shot publish
env2llm-mqtt publish --project . --probe-desktop

Topics (prefix env2llm):

Topic	Retain	Zawartość
{prefix}/{project}/registry	yes	full SystemMapIR JSON
{prefix}/{project}/registry/desktop	yes	desktop probe slice
{prefix}/{project}/events	no	refresh/metadata events
{prefix}/{project}/registry/refresh	—	subscribe: trigger refresh

Variable	Default
ENV2LLM_MQTT_ENABLED	off
ENV2LLM_MQTT_HOST	127.0.0.1
ENV2LLM_MQTT_PORT	1883
ENV2LLM_MQTT_TOPIC_PREFIX	env2llm
ENV2LLM_MQTT_USERNAME / PASSWORD	optional

Output formats

Format	File	Use case
doql.less	environment.doql.less	LLM + nlp2dsl orchestrator (default)
yaml	environment.yaml	Tooling, human edit
json	environment.json	APIs, CI
markdown	environment.md	Prompt injection summary

Environment variables

Variable	Purpose
NLP2DSL_BACKEND_URL	Gateway service URL
NLP2DSL_NLP_SERVICE_URL	NLP orchestrator URL
NLP2DSL_WORKER_URL	Worker executor URL
LLM_MODEL	LLM provider/model id
ENV2LLM_CONTEXT	Path to generated map (set after bootstrap)
ENV2LLM_DESKTOP_PROBE	1 to attach live desktop snapshot (same as --probe-desktop)
NLP2DSL_DOQL_CONTEXT	Alias for nlp2dsl compatibility

Secrets (*_KEY, *_TOKEN) are masked in output.

Extraction from nlp2dsl

This package contains the former nlp2dsl_sdk modules:

• doql/ — parse/render runtime context
• ir.py — SystemMapIR schema (env2llm.system_map.v1)
• generate.py, bridge.py, runtimes.py — introspection pipeline
• registry.py — live registry refresh after workflow steps
• policy/ — process and invoice policies from example-profiles.yaml

nlp2dsl can depend on env2llm and replace direct imports over time.

License

Licensed under Apache-2.0.