Local-first open-source operator workflow platform for self-hosted AI execution.

DreamAxis turns chat into a runtime-backed operator console for CLI, browser, and desktop work ¡ª with approval gates, audit trails, and rich final outputs instead of a hosted black box.

DreamAxis is built for operators and builders who want:

• no signup by default with AUTH_MODE=local_open
• self-hosted provider keys instead of a central account dependency
• OperatorPlan-backed execution instead of disconnected one-off prompts
• CLI + Browser + Desktop runtimes for real execution, not chat alone
• approval-gated desktop actions with runtime-backed audit and resume support
• proposal-only repo/code repair lanes instead of silent file edits
• rich final outputs with Markdown, code highlighting, KaTeX, and Mermaid rendering
• a stronger local baseline aligned with modern desktop coding assistants: Git + Node.js + pnpm/npm + Python

DreamAxis is designed around five simple defaults:

• local-first - run on your own machine and infrastructure first
• no-signup by default - the main path starts with AUTH_MODE=local_open
• operator-first - multi-step work is routed through plans, approvals, and visible execution state
• runtime-backed audit - chat, operator, and runtime stay linked to the same evidence trail
• self-hosted assets - keep provider keys, skills, knowledge, and workspace data under your control

Latest local acceptance baseline:

• v0.3.0-alpha.2 operator workflow: 9/9 scenarios passed across inspect desktop, verify browser, operate with approval, triad summary, failure reflection narrowing, and repo non-regression lanes
• rich text v1 rendering: fixed-fixture acceptance captured in 13 tracked screenshots across chat, operator, runtime, Mermaid fallback, HTML escaping, and narrow viewport coverage
• v0.2 chat-first repo copilot: 8/8 scenarios passed
• NVIDIA Build provider validation: 9/9 scenarios passed
• operator workflow core: multi-step OperatorPlan execution, approval queue handling, bounded reflection, deterministic desktop grounding, and resume are now validated end to end
• operator surfaces: /chat, /operator, and /runtime now divide cleanly into live operation, approval management, and deep audit
• rich output layer: final assistant and operator summaries now support Markdown, tables, code highlighting, KaTeX, and Mermaid without changing the backend message model
• Windows host desktop runtime: inspect, verify, and approval-gated operate flows validated with real focus_window, launch_app, press_hotkey, type_text, and click actions
• validated across:
  • DreamAxis
  • a Node.js repo
  • a Python repo
• focused on:
  • visible operator state
  • approval-gated desktop action
  • proposal-only repair output
  • runtime-backed evidence and parent/child execution linkage
  • rich final-message rendering

See:

• docs/acceptance-report-alpha2.md
• docs/acceptance/rich-text-v1/acceptance-report.md
• docs/chat-acceptance-report-v0.2.md
• docs/chat-acceptance-report-nvidia.md
• docs/repo-copilot-runbook.md
• docs/desktop-host-validation-2026-03-24.md

Operational overview for providers, runtimes, skills, knowledge, and workspace activity.

Skills	Operator	Runtime
Skill packs, execution entrypoints, and capability-aware actions.	Approval queue, active runs, templates, and plan-level controls for OperatorPlan-managed execution.	Runtime hosts, operator lineage, verification summaries, artifact-first audit, and child execution trails back to chat turns.

Knowledge	Chat
Builtin packs, uploaded documents, and retrieval-ready assets.	Operator-first chat console with active-step emphasis, approval prominence, runtime evidence, and rich final-message rendering for Markdown, code, math, and Mermaid.

See docs/screenshots.md for the canonical screenshot index and refresh rules.

Chat rendering	Code highlight	Math + KaTeX
Final assistant messages now support Markdown, links, lists, tables, and safe rich formatting.	Fenced blocks render with language labels, copy actions, and readable syntax highlighting.	Inline and block math render through KaTeX, including the fixture coverage used in acceptance.

Mermaid success	Mermaid fallback	Operator + Runtime rich text
Mermaid fenced blocks render client-side without changing the message model.	Failed Mermaid diagrams degrade locally with an error card and visible source block.	The same renderer now powers operator and runtime explanatory summaries while raw logs stay monospace.

DreamAxis now presents a clear alpha.2 shape:

• /chat = the live operator console
• /operator = approvals, queues, templates, and plan management
• /runtime = audit, lineage, artifacts, and execution detail

• bootstrap directly into the app with local_open
• use Docker for the full stack or run services separately
• validate machine and workspace readiness from /environment

• configure your own OpenAI-compatible base URL and API key
• sync available models or enter a model manually
• keep provider secrets self-hosted in your own deployment

• run CLI skills against your local workspace
• run Playwright-backed browser skills and capture artifacts
• inspect the Windows desktop surface through the host desktop runtime
• execute approval-gated desktop actions with runtime-backed audit trails
• manage OperatorPlan approvals and active runs from /operator
• follow active-step state in /chat and deep lineage in /runtime
• review runtime hosts, sessions, executions, and outputs in one place
• render final assistant and operator summaries with safe rich text instead of raw plain-text dumps
• use chat-first verify / troubleshoot flows with grounded targets, reflection-aware follow-up, and runtime-backed failure summaries instead of black-box answers

• default mode is AUTH_MODE=local_open
• no public registration flow is required for a local install
• metadata stays in your PostgreSQL
• uploaded knowledge files stay on your disk
• provider API keys stay self-hosted

DreamAxis already includes:

• CLI Runtime v1
• Browser Runtime v1 (Playwright)
• Desktop Runtime v1 (Windows host worker)
• OperatorPlan alpha.2 workflow layer with inspect / verify / operate / proposal sequencing
• runtime/session/execution visibility in the web console
• chat-first troubleshooting summaries and approval-gated desktop actions backed by runtime evidence, not prose-only diagnosis
• rich final-message rendering that upgrades explanatory output without changing the execution backend

• Builtin skill packs: core-cli, core-browser-playwright, core-research, core-docs, core-knowledge, core-repo
• Builtin knowledge packs: Playwright, Git, Docker, Python, TypeScript, FastAPI, Next.js, DreamAxis architecture
• OpenAI-compatible provider connections: user-supplied key, configurable base URL, dynamic model selection

DreamAxis treats the local environment as a product surface, not a hidden prerequisite:

• required: Git, Node.js, pnpm/npm, Python
• optional: Docker, Browser Runtime, Playwright
• Doctor page: checks readiness before a skill fails

See docs/environment-standard.md and docs/doctor.md.

Recommended local baseline:

• Git
• Node.js 22+
• pnpm 10+ or npm
• Python 3.12+
• Docker Desktop (recommended)

git clone https://github.com/DREAMVFIAUNION/dreamaxis.git
cd dreamaxis
pnpm install

Copy-Item .env.example .env

Recommended minimum:

AUTH_MODE=local_open
ENABLE_BROWSER_RUNTIME=true
JWT_SECRET_KEY=change-me-dreamaxis-development-secret
APP_ENCRYPTION_KEY=change-me-with-a-long-random-secret

docker compose -f infrastructure/docker/docker-compose.yml up --build

• Web: http://localhost:3000
• API health: http://localhost:8000/health

For the full development setup, non-Docker workflow, and reset instructions, see docs/development.md.

1. Enter directly with local_open
2. Open /settings/providers
3. Add an OpenAI-compatible API key
4. Sync models or enter one manually
5. Open /environment and confirm baseline readiness
6. Run one CLI skill
7. Run one Browser skill
8. Sync builtin knowledge packs
9. Upload a document
10. Open /chat/local-demo and send a knowledge-enabled message
11. Inspect /runtime for the execution trail

• user / workspace / provider / runtime / skill / knowledge metadata -> PostgreSQL
• provider API keys -> encrypted in provider_connections
• uploaded documents -> KNOWLEDGE_STORAGE_PATH
• browser auth token -> local browser storage

DreamAxis does not require a hosted account system for the default path.

• /dashboard
• /chat/[conversationId] - live operator console with approval-aware execution state and rich final outputs
• /operator - approval queue, active runs, templates, and plan management
• /skills
• /knowledge
• /runtime - audit plane with lineage, artifacts, raw logs, and execution detail
• /environment
• /settings/providers
• /acceptance/rich-text-v1 - fixed-fixture rendering acceptance harness
• /login (only for optional password mode)

• docs/architecture.md
• docs/development.md
• docs/deployment-modes.md
• docs/browser-runtime.md
• docs/acceptance-report-alpha2.md
• docs/desktop-runtime-v1.md
• docs/acceptance/rich-text-v1/acceptance-report.md
• docs/skill-packs.md
• docs/knowledge-packs.md
• docs/backend-api.md
• docs/skill-requirements.md
• docs/launch/README.md
• docs/launch/csdn-build-log-tutorial.md
• docs/v0.3-desktop-operator-plan.md
• ROADMAP.md
• CHANGELOG.md

• CONTRIBUTING.md
• CODE_OF_CONDUCT.md
• SECURITY.md
• SUPPORT.md

DreamAxis is released under the MIT License.