Private preview for local-first AI operators

Turn intent into
finished work on your own machine.

Ferryman is a local-first desktop AI workspace built around Sessions, Skills, Tasks, Artifacts, Runtime Logs, Model Management, and Model Routing.

Private preview for teams that need dependable automation, not another chat box.

Operational proof, not AI theater

Every run leaves a trail: context lives in Sessions, execution appears in Tasks and Runtime Logs, outputs land in Artifacts, and model choices stay visible.

Local-first BYOK Private Preview
01 · Session

Each project keeps its own context, history, and working memory instead of dissolving into one endless chat.

AI Hotspot Daily Brief Live
geosolver.app keyword research Saved
02 · Task Monitor

Execution is broken into visible tasks so you can see what is running, waiting, finished, or blocked.

Distill target pages running
ReadSkillSOP done
Write report queued
03 · Skill

A Skill is an operational playbook: SOP, tool scope, working assumptions, and output shape bundled into one reusable unit.

SOP Tools Artifacts
skill: seo-keyword-research
tools: browser, read_file, write_file
deliverable: keyword-map.md
04 · Artifacts

Ferryman writes outputs to disk so the result is a usable file, not just text that disappears in chat history.

/artifacts/seo-brief.md
/artifacts/source-notes.json
/artifacts/final-outline.md
05 · Runtime Logs

When something breaks, the trail is still there. Logs turn debugging from guesswork into diagnosis.

21:08:14 kernel.connected session=campaign-q2
21:08:19 tool.browser.navigate status=ok
21:08:26 artifact.saved path=seo-brief.md
06 · Model Management

Providers, keys, base URLs, and active models stay in your control instead of being hidden behind one vendor opinion.

OpenAI linked
Gemini active
Current model: gemini-2.5-pro
07 · Model Routing

Different work deserves different engines. Ferryman can route planning, execution, and formatting to the model path that fits the job.

Plan with a reasoning model session → planning
Execute with a faster worker task → tool calls
Finalize into deliverables artifact → saved output

Why it holds up

Built for work that
actually has to finish.

Ferryman is not trying to look magical. It is designed to stay legible when execution, credentials, files, and failure states matter.

Local-first execution

Keep tasks, outputs, provider settings, and browser work close to the operator instead of pushing everything into a remote black box.

Visible execution

Sessions, task state, token usage, runtime logs, and saved artifacts make the system understandable while it works.

Model control and routing

Bring your own providers, choose the active model, and route different stages of work through the model path that makes sense.

Human handoff when needed

Ferryman is designed for real browser work, so the system can surface checkpoints instead of pretending every edge case should be automated away.

Skills

A Skill is not a prompt. It is an operating playbook.

Each Skill packages process, tools, assumptions, and output shape so the same intent can be executed repeatedly without turning into vague chat instructions.

A · SOP and guardrails

A Skill defines what good execution looks like before a model starts improvising. That means scope, steps, constraints, and expected deliverables are already shaped.

goal: produce a concise operator-ready deliverable
constraints: no invented sources, write outputs to artifacts
handoff: ask the operator when judgment is required
B · Tools, memory, and outputs

A Skill is tied to concrete tool access and expected outputs. It can read files, drive the browser, write artifacts, and carry local context without pretending the world is just text.

Browser toolsrisc
Local files and artifactswrite
Session memory and task tracetrace
C · What the operator gets

The result is a durable working surface, not a one-off answer.

brief.md
sources.json
next-actions.md

Ferryman is opinionated about execution so your workflows stay reusable.

System flow

Intent enters. Deliverables leave.

1

Session holds the context

Requests do not float in isolation. Sessions preserve history, task state, and the working surface for a given stream of work.

2

Routing chooses the path

Ferryman decides which Skill, which tool surface, and which model path should carry the request forward.

3

Execution stays legible

Browser actions, file writes, task updates, and runtime logs remain inspectable while the system is moving.

4

Artifacts make the run durable

The end state is a saved output you can inspect, hand off, version, or reuse, not a transient answer lost in scrollback.

Flow
01
Intent enters a Session
02
Model Routing chooses the reasoning path
03
Skill SOP defines execution shape
04
Browser, files, and tools run locally
05
Tasks and Runtime Logs stay observable
06
Artifacts are written back to disk

Need AI that can
finish the work?

Ferryman is currently available by private preview.