InsightWorker Logo
  • contact@verticalserve.com

Download InsightWorker

Free trial through 31 August 2026 · CLI · Desktop · 5 LLM providers

FREE TRIAL · ENDS 31 AUG 2026

Pick your install

The CLI is for engineers and SREs. The Desktop app is for analysts and operations teams. Same agent core, same skills.

For business users

Desktop App

Folder browser, drag-and-drop into chat, app workspaces, in-app permission manager, multi-provider LLM settings. New in v2.2.0: §15 app bundle authoring, Publish to S3, "Open in InsightStudio", and a one-click --worker mode to join a Studio coordinator's farm.

Latest: v2.2.0 — macOS Apple Silicon & Intel. Windows + Linux on the next release.

Windows installer is currently on v2.0.3 — v2.2.0 builds for Windows + Linux ship next release.

macOS — first run: The trial builds are unsigned. After dragging InsightWorker to Applications, open Terminal and run:
xattr -cr /Applications/InsightWorker.app
Then double-click to open. The "is damaged" error you may see otherwise is macOS Gatekeeper blocking unsigned + quarantined apps — the file itself is fine.
Windows — first run: SmartScreen may say "Windows protected your PC" → click More infoRun anyway.
For engineers / SREs

Command-line

Interactive REPL, one-shot mode for CI, slash commands, project workspaces. The install script handles the Node 20 LTS runtime for you (or bring your own — Node 18+ works). Four flavors of the same install: REPL, browser UI (--web), per-app widget form (--ui), or worker daemon that joins an InsightStudio farm (--worker).

Latest: v2.1.11 — ships the worker daemon and per-run artifact downloads.

curl -fsSL https://insightworker.com/install.sh | bash

After install, try any of:

insightworker                            # interactive REPL
insightworker --web                      # browser UI on 127.0.0.1:8765
insightworker /app create "<name>" --ui    # widget form for one app
insightworker --worker --studio <url> \
              --token <bearer>           # join an InsightStudio farm

The --worker mode turns this box into a daemon that polls an InsightStudio coordinator and runs queued jobs from its browser-based marketplace. See Join a worker farm for the full flag reference and a systemd unit.

Or download the tarball: insightworker-latest.tgz

Linux & macOS supported today. Windows CLI installer is on the next release.
For developers using VS Code, Cursor, Kiro, or VSCodium

InsightWorker — Editor Extension

Drop InsightWorker into the editor you already use. Adds a chat panel, command-palette tools, and a status-bar provider indicator. The agent runs as a subprocess of the CLI you've already installed — no separate auth, no extra service.

After downloading, install with one command (works in VS Code, Cursor, Kiro, and VSCodium):

code --install-extension ~/Downloads/insightworker-vscode-latest.vsix

(Or in the editor: Command PaletteExtensions: Install from VSIX… → pick the file.)

Prerequisite: the InsightWorker CLI must be installed and configured first (the extension wraps the CLI). Run the install line above, then add the extension.

What you get:

  • Chat panel in the activity bar — same agent, same tools (SharePoint, JIRA, Perplexity, databases, Outlook), inline tool-call visibility
  • Command palette: InsightWorker: New Chat, Run One-Shot Prompt, Open ~/.insightworker/.env
  • Permission prompts surface as inline Allow / Deny buttons in the chat panel
  • Status bar shows your active LLM provider · model
Works on macOS (Intel + Apple Silicon), Windows, and Linux — same .vsix, same install command in any VS Code-family editor.
For organizations · 10+ seats

Need centralized control across your team?

InsightWorker Enterprise adds SSO, per-team model whitelists, token quotas, a private skills marketplace, and audit logging on top of the same product. Self-hosted or SaaS.

Configure your LLM provider

InsightWorker is provider-agnostic. Set one of the following in ~/.insightworker/.env (CLI) or in Settings → LLM Provider (Desktop).

AWS Bedrock
Anthropic API
OpenAI
Azure OpenAI
Custom / GPU
Google Gemini
Vertex AI

AWS Bedrock — recommended for enterprise

Uses your existing AWS credentials (~/.aws/credentials or IAM role). No model API key to manage.

1
Enable model access in the AWS Console. Open Bedrock → Model access in your region and request access for the Anthropic Claude family.
2
Verify with the AWS CLI that the model you want is ACTIVE:
aws bedrock list-foundation-models --region us-east-1 \
  --by-provider anthropic \
  --query 'modelSummaries[?modelLifecycle.status==`ACTIVE`].modelId'
3
Configure InsightWorker. Edit ~/.insightworker/.env:
LLM_PROVIDER=bedrock
AWS_REGION=us-east-1
BEDROCK_MODEL=us.anthropic.claude-sonnet-4-5-20250929-v1:0

# Optional — falls back to ~/.aws/credentials or IAM role
# AWS_ACCESS_KEY_ID=...
# AWS_SECRET_ACCESS_KEY=...
If you see "This Model is marked by provider as Legacy", replace BEDROCK_MODEL with an ACTIVE ID from step 2.

Anthropic API — direct

Use this if you don't have AWS Bedrock and want to call Claude directly via Anthropic's API.

1
Get an API key from console.anthropic.com (starts with sk-ant-).
2
Edit ~/.insightworker/.env:
LLM_PROVIDER=anthropic
ANTHROPIC_API_KEY=sk-ant-...
AGENT_MODEL=claude-sonnet-4-5
Use caseRecommended model
General appsclaude-sonnet-4-5
Stored-proc translation, complex reasoningclaude-sonnet-4-5
Bulk classification, cheap callsclaude-haiku-4-5

OpenAI

Use a key from platform.openai.com (starts with sk-).

1
Create a project key with access to the model you want.
2
Edit ~/.insightworker/.env:
LLM_PROVIDER=openai
OPENAI_API_KEY=sk-...
OPENAI_MODEL=gpt-4o
Use caseRecommended model
General apps, tool usegpt-4o
Cheap classification / draftsgpt-4o-mini
Hard reasoning, planningo1-preview · o1-mini
Tool-call arguments come back as JSON strings. InsightWorker parses them defensively and reports a clean error if a model returns malformed JSON.

Azure OpenAI

For enterprise deployments where prompts must stay in your Azure tenant. Same Chat Completions surface as OpenAI proper, routed through your endpoint and deployment.

1
In the Azure portal, create an OpenAI resource and a model deployment (e.g. my-gpt-4o).
2
Edit ~/.insightworker/.env:
LLM_PROVIDER=azure
AZURE_OPENAI_ENDPOINT=https://my-resource.openai.azure.com
AZURE_OPENAI_API_KEY=...
AZURE_OPENAI_DEPLOYMENT=my-gpt-4o
AZURE_OPENAI_API_VERSION=2024-08-01-preview
The deployment name takes the place of the model id. The API version defaults to 2024-08-01-preview — bump it when you upgrade Azure resources.

Custom OpenAI-compatible (vLLM, Ollama, LM Studio, in-house GPU)

Point at any endpoint that exposes the OpenAI Chat Completions API. Useful for self-hosted models, local dev, or your own GPU box.

1
Stand up the endpoint. Examples:
# vLLM
vllm serve meta-llama/Llama-3.1-70B-Instruct --port 8000

# Ollama (already exposes /v1)
ollama serve

# LM Studio — enable "OpenAI-compatible server" in Local Server tab
2
Edit ~/.insightworker/.env:
LLM_PROVIDER=custom
CUSTOM_LLM_BASE_URL=http://gpu-box.internal:8000/v1
CUSTOM_LLM_API_KEY=optional-if-required
CUSTOM_LLM_MODEL=meta-llama/Llama-3.1-70B-Instruct
EndpointTypical base URL
vLLM / TGIhttp://<host>:8000/v1
Ollamahttp://<host>:11434/v1
LM Studiohttp://localhost:1234/v1
Prompts and tool outputs are sent to whatever URL you configure. Only point at endpoints you trust. The base URL is never auto-set — you must opt-in by setting the env var.
The model must support OpenAI-style tool calling for apps to work end-to-end. Most modern instruct-tuned models do (Llama 3.1+, Qwen 2.5+, Mistral Large). Older models may chat fine but fail when asked to use tools.

Google Gemini (AI Studio)

Use a key from aistudio.google.com/apikey. Fastest path — no GCP project required.

1
Create an API key in AI Studio (starts with AIza).
2
Edit ~/.insightworker/.env:
LLM_PROVIDER=gemini
GEMINI_API_KEY=AIza...
GEMINI_MODEL=gemini-2.0-flash-001
Use caseRecommended model
General apps, tool usegemini-2.0-flash-001
Hard reasoning, long contextgemini-2.5-pro
Cheap classificationgemini-2.0-flash-lite
Gemini's function-calling schema is OpenAPI 3.0-ish — stricter than full JSON Schema. InsightWorker sanitizes tool schemas automatically (strips $ref, oneOf, etc.). If a custom tool fails on Gemini but works on OpenAI, simplify its input_schema.

Google Vertex AI

For enterprise deployments where prompts must stay in your GCP project. Same Gemini models, IAM-governed, region-pinned, with VPC-SC support.

1
Enable the Vertex AI API in your project. Grant your principal the roles/aiplatform.user role.
2
Auth via Application Default Credentials (run gcloud auth application-default login) or point at a service-account JSON.
3
Edit ~/.insightworker/.env:
LLM_PROVIDER=vertex
GOOGLE_CLOUD_PROJECT=my-gcp-project
GOOGLE_CLOUD_LOCATION=us-central1
# Optional — falls back to ADC if unset:
GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
VERTEX_MODEL=gemini-2.0-flash-001
Common locations: us-central1, us-east5, europe-west4, asia-northeast1. Some Gemini models are only available in a subset of regions — check the Vertex AI locations doc.

Live web search

Plug in a Perplexity Sonar key and the agent gets a perplexity_search tool — current news, market data, regulatory filings, all with source citations.

Setup

Get a key at perplexity.ai/settings/api, then add to ~/.insightworker/.env:

PERPLEXITY_API_KEY=pplx-...
PERPLEXITY_MODEL=sonar-pro

The agent decides when to call it. Sample prompts that route through Perplexity:

  • "Any material updates about Apple in the last hour?"
  • "What did the Fed announce at this morning's meeting?"
  • "Find regulatory filings for Snowflake from the past week."

Tool input supports recency (hour / day / week / month / year) and domain_filter (allowlist or -domain to exclude).

Output is auto-truncated to the active LLM's context budget — Gemini gets up to 800k chars, gpt-4o gets ~50k, so multi-turn agent loops don't blow past the model's window.

First run

Get to a working chat in under two minutes.

CLI

Start the REPL

insightworker
> what DAGs are running in Airflow?
> compare row counts MySQL ↔ Oracle for users
> /app create "convert Redshift DAG to Oracle ADW"
Desktop

Open a folder, ask a question

  1. Launch InsightWorker from Applications.
  2. Click + Open Folder… and pick a project folder.
  3. Drag any file from the Files tab onto the chat to attach.
  4. Use Apps → + New for repeatable tasks.

Schedule recurring runs

Drop a YAML file in your workspace, run insightworker --daemon, and the agent fires on a cadence — every minute through every day. Pair with the Perplexity tool to monitor companies, markets, or systems on autopilot.

Schedule file

<workspace>/.insightworker/schedules.yaml:

schedules:
  - name: monitor-apple
    every: 1h
    prompt: |
      Use perplexity_search recency=hour for "Apple Inc news".
      If anything material (product launch, M&A, exec change,
      earnings, big stock move) appeared in the last hour,
      append a dated entry to ~/Desktop/apple-watch.md.
      If nothing material, skip the write.

  - name: nightly-cleanup
    every: 24h
    app: cleanup-stale-dags
    enabled: true

Cadence syntax: 1h, 30m, 1h30m, 45s, @hourly, @daily, @weekly. 30-second floor as a rate-limit guard. Use prompt: for free-form text or app: to point at an existing app under .insightworker/workflows/.

CLI
# Run the scheduler in foreground
insightworker --daemon

# Inspect what's loaded
insightworker schedule list

# Fire one immediately (skips cadence)
insightworker schedule run monitor-apple
Production wrap

For unattended runs, wrap with launchd (macOS), systemd (Linux), or pm2. Each schedule writes to its own log:

~/.insightworker/logs/<schedule>.log

Per-schedule mutex skips overlapping runs (won't stack a second tick on top of an in-flight one). SIGINT / SIGTERM wait for in-flight runs to finish before exit.

Trial license

This build runs free through 31 August 2026. After that, the CLI and Desktop app refuse to start until a license key is configured. To purchase, extend, or move to enterprise tier:

Contact our team

contact@verticalserve.com