Skip to content

Choose how you use AI

Install Occam, choose one workflow, and finish with a real page read. Do not stop at “installed.”

Quick connection test (same for every path):

Use Occam to read https://example.com/ and summarize the page. Include the source URL.

This proves the tool call works. The page is deliberately small.

Optional richer prompt (after the connection test succeeds):

Use Occam to read https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide and summarize what the guide covers. Include the source URL.

This is a stable public documentation page used in Occam’s smoke corpus — useful enough to show compact, source-linked content without relying on a flaky site.

The recorded first useful job (MDN Functions, budget + listed omissions) is in the workflow gallery. Host-specific where-to-type: onboarding.

Choose your path

Cursor Validated New Cursor chat after connect Open Cursor guide → Hermes Agent Supported · verification required Only if Hermes is genuinely installed Open Hermes guide → Another AI app Support varies Another MCP-capable application Use another AI app → Local Ollama model Experimental Type in the Occam chat terminal Start local Ollama chat →
No supported app detected?

The install can still succeed. Connect a host later, use experimental chat, or integrate your own MCP client.

Open recovery options →

Occam normally appears as a tool inside your AI application. The occam terminal commands install, connect, check, and diagnose that tool. The experimental Ollama path is different: you type directly in an occam chat terminal.

Install once

Requires Node.js 20+. The release install does not require the .NET SDK.

curl -fsSL https://raw.githubusercontent.com/ContextForgeAI/occam/main/scripts/get-ff-occam.sh | bash

irm https://raw.githubusercontent.com/ContextForgeAI/occam/main/scripts/get-ff-occam.ps1 | iex

The currently published channel is v1.1.1. The installer verifies the release archive against its SHA-256 manifest, verifies Cosign when signaturePolicy=required-cosign-v1 is declared (requires the cosign CLI), runs archive-member preflight before extract, and checks the complete bundled runtime from that archive before replacing an install — without fetching post-install helpers from the mutable repository branch. It then checks the local runtime, detects supported applications, and explains the next action. If it detects one supported application, it can connect it. If it finds several, it asks which one to configure.

Release binaries exist for Windows x64, Linux x64, and Apple Silicon macOS. An unsupported CPU/OS combination is rejected before download.

Read the final status before continuing:

Connected and readyThe host confirmed the connection
Needs restartConfiguration is present; restart or reload the named application
Not connectedThe attempt failed verification or was undone; Occam is not available in that application
Needs your actionComplete the named trust, permission, conflict, or paste step
No apps readyOccam is installed, but no AI application is connected yet

npm / npx is not the supported 1.0 release install path. See Install for the full contract.

Path A: supported AI application

Use this path for a supported application that the installer reports as ready, configured, or awaiting restart. Validation tiers and exact behavior differ by host; see MCP hosts.

  1. Install Occam.

  2. Select an application when the installer asks, or run:

    occam connect

  3. Follow the exact restart, reload, trust, or paste instruction in the result.

  4. Open a new conversation in the selected application.
  5. Type the quick connection test prompt.
Where do I type?
In a new conversation in the connected AI application.
What should happen?
The application asks an Occam page-reading tool to read the live URL, then summarizes the returned page content.
How do I know Occam was used?
If the application shows tool activity, look for an occam_* call, usually occam_transcode. The result should identify https://example.com/ as its source. Do not rely on a plausible answer alone.
What if it was not used?
Run occam status, then occam smoke. If the local host is healthy, run occam connect again and follow its host-specific instruction.

Success signal

A real tool call and live page content — not the model explaining what Occam is or running which occam in a shell.

Path B: Cursor

Cursor has a validated connection path, but a configuration write does not make an already-open chat reload its tools.

  1. Install Occam.

  2. If Cursor was not connected during install, run:

    occam connect --only cursor

  3. Restart or reload Cursor if the connect result asks.

  4. Open a new Cursor chat.
  5. Type the quick connection test prompt.
Where do I type?
In a new Cursor chat after the requested restart or reload.
What should happen?
Cursor calls an Occam tool and uses the returned Example Domain content.
How do I know Occam was used?
Inspect the tool activity Cursor exposes for that chat and look for an occam_* tool. UI labels can change, so this guide does not depend on a specific button name.
What if tools are absent?
Run occam smoke. If it passes, run occam connect --only cursor, follow the printed restart/reload step, and create another new chat.

Path C: Hermes Agent

Hermes must be genuinely installed. A leftover configuration directory is not enough, and configuration alone does not prove that Hermes loaded Occam. Continue only when connect verification succeeds — do not treat Hermes as ready from a config write alone.

  1. Install Occam.

  2. Connect Hermes:

    occam connect --only hermes

  3. Continue only if the result says Hermes is ready or gives a concrete restart/session instruction. If the result is Not connected, stop and use another path.

  4. Open a new Hermes conversation.
  5. Type the quick connection test prompt.
Where do I type?
In a new Hermes conversation, not in the terminal used for occam connect.
What should happen?
Hermes calls the local Occam MCP server and summarizes the live page.
How do I know Occam was used?
Hermes should expose or call Occam tools. A shell lookup such as which occam is not an Occam web read.
What does Not connected mean?
Verification did not establish readiness. Occam may have undone the attempted registration. Do not test Hermes as if Occam were available.
What should I retry?
Run occam doctor, then occam connect --only hermes. Follow the resulting diagnostic or use another validated path.

Path D: local Ollama model (experimental)

occam chat is an experimental local-chat path, not a stable 1.0 interface. Ollama is a model runtime, not an MCP host.

  1. Start Ollama.
  2. Make sure at least one installed model reports tool support.

  3. Run:

    occam chat

  4. Select a tool-capable model or accept the offered default.

  5. At the > prompt, type the quick connection test prompt.
  6. Type /exit to leave the chat.
Where do I type?
In the terminal opened by occam chat.
What should happen?
The local model asks Occam to read the URL, and the terminal shows a reading/tool status before the answer.
How do I know Occam was used?
The answer follows an Occam tool call and uses the live Example Domain content.
What if it was not used?
Read the terminal message. Start Ollama if it is unreachable; install/select a tool-capable model if no compatible model is available; run occam doctor if the Occam tools are unavailable.

This path requires no Ollama login and does not use Ollama Web Search. Occam calls the documented local Ollama API and uses its own acquisition stack. It is not a native integration inside the Ollama App. A custom remote OLLAMA_HOST changes the network boundary.

No model names are promised here: compatibility is experimental and should be checked against the models installed on your machine.

Path E: no supported application detected

No detected host does not mean the Occam installation failed. It means the local runtime is installed but has nowhere to expose its tools yet.

Choose one:

  1. Install a supported AI application, then run:

    occam connect

  2. If local Ollama is running with a tool-capable model, use the experimental chat path.

  3. If you are integrating your own MCP client, generate the launch snippet:

    occam snippet

    Then use the transport and MCP API documentation.

Where do I type the first prompt?
In the supported application you connect later, the occam chat terminal, or your custom MCP client.
What exactly do I type?
Use the same quick connection test at the top of this guide.
What should happen?
The chosen host completes a real Occam tool call and returns source-linked page content.
What if it does not?
Run occam status, occam doctor, and occam smoke; then connect or configure the chosen host.

Diagnostics

These are current public commands:

occam status   # installed version and setup summary
occam doctor   # runtime, browser, and web-safety checks
occam smoke    # local MCP initialize, tools/list, and live probe
occam connect  # detect and connect supported applications
occam update   # read-only release check

Use the result in this order:

  1. If occam doctor fails, fix the local runtime first.
  2. If occam smoke fails, the local MCP host is not ready.
  3. If both pass but the application has no tools, reconnect that application and follow its restart/reload instruction.
  4. If connect reports Not connected, do not treat that host as ready.

More help: Troubleshooting.

Preview removal

Before deciding to keep Occam, you can inspect exactly what removal would change:

occam disconnect --dry-run  # managed host registrations only
occam uninstall --dry-run   # registrations, generated launcher, release tree

Default uninstall preserves local state, skills, backups, the response cache, and the shared Playwright browser cache. State and cache cleanup require explicit flags. See installation safety: disconnect and uninstall for the complete boundary.

You have your first result when

First-result checklist

  • your selected application or chat terminal made a real Occam tool call;
  • the result contains the live Example Domain page content and source URL; or
  • Occam returned ok: false and the application handled the page as unknown instead of inventing content.

The reproducible success and controlled failure are available in the current proof fixture.

After first success:

Read a difficult pageJavaScript-heavy sources Authenticated accessSessions / login walls Research several URLsMulti-source digest Check an integrity receiptVerify a source Understand the architectureHow Occam works