Private assistant workspace
Run internal models, use tool-enabled answers, share read-only conversations, and preserve retrieval and metadata context.
Pulsar
Spaceport Consulting
Private deployment, full AI workspace
Pulsar: private AI, deployed on your terms.
Pulsar brings private chat, scheduled work, image and video generation, a repository-aware coding agent, asset library, and operator controls into one deployable workspace.
Runs inside your environment
Deploy behind your proxy, on your network, with your policies.
Bring your own models
Use vLLM, LM Studio-compatible endpoints, Ollama, and ComfyUI.
Agentic coding
Pulsar Code reads repos, prepares changes, runs checks, and gates PRs.
A complete private AI workspace, not a single chat box.
Pulsar organizes the surfaces teams actually use: conversation, scheduled tasks, generated media, repository-aware coding, assets, tools, and administration.
Recurring AI work
Create one-time or recurring Chat, Pulsar Code, and monitor tasks with visible history and linked outputs.
Media generation suite
Create, edit, inspect, and manage image and video jobs through a consistent FLUX-aware workspace.
Pulsar Code
Pick a repository, describe the goal, review evidence and the proposed patch, then approve branch and pull-request actions.
Asset hub
Find generated images, videos, files, Pulsar Code reports, exports, and saved answers with consistent item actions.
Control room
Monitor services, queues, logs, diagnostics, users, billing, storage, runners, and product configuration.
Runtime flexibility
Use the model stack that fits your deployment.
Pulsar is not tied to one hosted model vendor. It can run against GPU-backed inference, desktop/lab OpenAI-compatible servers, CPU utility models, and ComfyUI media workflows.
vLLM
Use vLLM as a private model runner for high-throughput GPU inference, longer context, and production OpenAI-compatible chat endpoints.
LM Studio
Connect LM Studio-compatible endpoints as another model runner path for local labs, smaller deployments, demos, and model testing.
Ollama
Run CPU-friendly utility models for prompt optimization, chat titles, summaries, diagnostics, and safe self-healing recommendations.
ComfyUI
Power image creation, image edit, gallery, and video workflows while admins control model assets, LoRAs, queues, and fallback behavior.
Coding agent
Describe the goal. Pulsar Code handles the repo work.
Pulsar Code is Pulsar's repository-aware coding workspace. The user sees a simple task thread and review surface while containers, repository policy, file evidence, diffs, checks, approvals, and diagnostics stay available underneath.
- Read and explain
Ask about a repo, file, architecture choice, bug, or test failure. - Prepare changes
Generate a patch, inspect the diff, and keep PR/write actions approval-gated. - Review safely
See files read, files changed, checks, and version comparison without exposing internal runner details by default.
Media and assets
Image, video, and saved work share one product language.
Pulsar brings creation, editing, galleries, video jobs, uploaded files, generated reports, and downloaded exports into a consistent workspace rather than scattering them across disconnected tools.
- FLUX-aware generation
Create and edit images with engine-compatible controls and admin-governed defaults. - Generated video workflows
Use image-to-video or text-to-video paths with queue status and library retention. - Asset actions
Open, download, inspect metadata, copy links, or delete items where allowed.
Library keeps outputs findable.
Images, videos, files, Pulsar Code reports, exports, and saved answers live in a single asset hub.
Tools are managed, not hidden.
Enable or review capabilities like Fetch, Brave, GitHub, Image, Video, Pulsar Code, and API/MCP access.
Operations
A control room for self-hosted AI.
Pulsar treats operations as a first-class product surface. Admins can see service health, queues, logs, diagnostics, storage, databases, runners, image settings, utility models, users, billing, and governance from one place.
- Runtime visibility
Track vLLM, LM Studio endpoints, Ollama utility runner, and ComfyUI alongside Valkey, Postgres, SeaweedFS object storage, and app services. - Diagnostics and safe recovery
Analyze logs with redaction, review incident cards, and run predefined safe actions. - Governance
Manage users, roles, MCP access, product capabilities, billing states, and storage policy.
Built for private teams that still need momentum.
Every major surface follows the same pattern: ask in plain language, inspect the result, and keep high-risk actions gated.
Choose the surface
Chat, Scheduler, Images, Pulsar Code, Library, Tools, or Admin.
Describe the outcome
Use a prompt, recurring task, repository goal, media request, or operator action.
Review evidence
Inspect linked references, files, outputs, logs, diffs, checks, and history.
Approve what matters
PRs, writes, destructive actions, and config changes stay explicit.
Documentation
Documentation built from the same runbooks used to operate Pulsar.
This high-level guide maps the private repo documentation into the areas evaluators need first: install shape, runtime choices, product workflows, and release operations.
Install and deployment
Bring up the Pulsar stack with Docker services, storage, reverse proxy routing, and health checks.
- App host, compose profiles, Valkey, Postgres, SeaweedFS object storage, and workers.
- Nginx and upstream proxy shape for controlled networks.
- Sanity checks, update flow, backups, and production hardening.
Runners and models
Choose how Pulsar connects to private model runners for chat, utility work, and media workflows.
- vLLM and LM Studio-compatible endpoints are both supported model runner paths.
- vLLM fits GPU-backed production inference; LM Studio fits local labs and smaller deployments.
- Optional Ollama utility runner handles titles, summaries, diagnostics, and self-healing recommendations.
Chat, Scheduler, and Library
Understand the core user workspace: conversations, recurring work, saved assets, and output history.
- Private chat with tools, shared read-only links, and persisted metadata.
- Scheduled chat, Pulsar Code, and monitor tasks with run history.
- Library categories for files, media, reports, saved outputs, and exports.
Pulsar Code
Operate the repository-aware coding agent without exposing the underlying container and policy complexity.
- Pick a repository, describe a goal, and review task results.
- Inspect file evidence, diffs, checks, and prepared patches.
- Keep writes and PR creation behind approval gates.
Images and video
Configure generation and editing workflows while keeping runtime-specific controls understandable.
- ComfyUI, FLUX-native generation, and explicit SDXL fallback paths.
- Image create, edit, video, gallery, LoRA compatibility, and queue recovery.
- Prompt optimization, reference guidance, and media retention.
Admin and release operations
Run Pulsar as an operator with clear visibility into services, safety gates, and promotion readiness.
- Control room, diagnostics, logs, queues, storage, users, billing, and audit trail.
- QA release gate, sanity checks, known risk review, and production guardrails.
- Safe self-healing for stale jobs, scheduler cleanup, runtime cache refresh, and known-safe retries.
Release candidate
RC availableDownload Pulsar for your environment.
Use the same immutable Pulsar release on Windows 11 with WSL or native Ubuntu. Each installer verifies the candidate manifest and deployment bundle before setup begins.
Windows 11
Windows + WSL installer
Sets up Ubuntu 24.04 in WSL, the Pulsar runtime, launch shortcuts, and hardware-aware model options.
Download for Windows
Install on Windows
Use Windows 11 with hardware virtualization enabled, an administrator account, internet access, and enough free storage for the models you select.
- Download the installer, then open PowerShell as Administrator in your Downloads folder.
- Verify the release checksum and run the unsigned installer with a one-process policy bypass:
$file = ".\Pulsar-Setup-db8ea54202b411d688c10c46e058fb8c57b1f808.ps1"
if ((Get-FileHash $file -Algorithm SHA256).Hash -ne "478638498993D5B3D570F1B71552DE8CD57A534F74A9CFFE5CE928773C12419E") { throw "Checksum mismatch" }
Unblock-File $file
powershell.exe -NoProfile -ExecutionPolicy Bypass -File $file- If Windows asks for a restart after enabling WSL, restart and resume from the same folder:
powershell.exe -NoProfile -ExecutionPolicy Bypass -File ".\Pulsar-Setup-db8ea54202b411d688c10c46e058fb8c57b1f808.ps1" -Action Resume- Accept the displayed Pulsar and model terms, then complete the hardware, model, and optional media prompts. Keep the terminal open while images and models download.
- Open Pulsar from the Windows Start menu. The shortcut starts the WSL services and opens Pulsar in your browser.
The policy bypass applies only to the installer process. It does not change the system execution policy.
Ubuntu 22.04 / 24.04
Native Ubuntu installer
Runs the interactive CPU, NVIDIA, or AMD setup and preserves install state across resume and upgrade.
Download for Ubuntu
Install on Ubuntu
Use an x86_64 Ubuntu 22.04 or 24.04 host with sudo access, internet access, and enough free storage for the models you select.
- Open a terminal. Download the immutable RC installer directly with
curl. - Verify its checksum, then start the interactive installer:
cd ~/Downloads
curl -fLO "https://github.com/dbesade/pulsar-downloads/releases/download/rc-db8ea54202b411d688c10c46e058fb8c57b1f808/Pulsar-Setup-db8ea54202b411d688c10c46e058fb8c57b1f808.sh"
echo "3d43331cd831c31ca8f28d165d1e74284daf9a85fe96e833b2bd9c7c0531f26a Pulsar-Setup-db8ea54202b411d688c10c46e058fb8c57b1f808.sh" | sha256sum -c -
sudo bash Pulsar-Setup-db8ea54202b411d688c10c46e058fb8c57b1f808.sh- Choose the detected CPU, NVIDIA, or supported AMD configuration, then select primary, utility, and optional media models.
- Keep the terminal open while Pulsar downloads and verifies runtime images and model assets.
- When setup completes, open http://localhost:3000 on the host or http://<server-ip>:3000 from a trusted network.
No GitHub account or container-registry login is required. A Hugging Face token is requested only when a selected model requires acceptance or authenticated download.
Downloads and runtime images are anonymous. This RC is unsigned; verify it with SHA256SUMS.
Private deployment consultation
Bring us the environment. We will map the right Pulsar deployment.
Tell us what needs to stay private, where your models run, and what your team wants to accomplish. The request goes directly into the Pulsar relationship workspace for structured follow-up.
Architecture first
We start with infrastructure, security boundaries, model runners, and operational ownership.
One accountable thread
Your request, context, owner, stage, notes, and next action stay together.
No public-cloud assumption
Pulsar is designed around controlled environments and deployment-specific constraints.