Skip to content

Releases: matank001/clodfarm

clodfarm 1.13.0: an X connector, and your computer's sessions on the farm

Choose a tag to compare

@matan-duke matan-duke released this 11 Oct 18:54

1.13.0 (2026-10-11)

  • An X connector, through X's own API. The farm manager connects the farm's X developer app once: CONNECTORS → X,
    its API key and secret and an access token and secret for the account (OAuth 1.0a, from the developer console's
    Keys and tokens), and the bearer token if there is one. The farm checks them with GET /2/users/me, signs every
    call itself, and keeps them in .farm/connectors/x.json. Every Claude gets clodfarm x: me, user, posts,
    mentions, search, get, post TEXT [--reply-to] [--quote] [--media FILE ...], delete, and api METHOD PATH
    for any other v2 call. Its guide says to post, reply, like, follow, DM or delete only when its person asks, and to
    keep reads small, since X bills each call. A read-only token says so on the card. From a shell:
    clodfarm x connect (JSON on stdin) and clodfarm x disconnect. See docs/connectors.md#x.
  • Your computer's own Claude Code sessions, on the farm: clodfarm attach. Until now only the farm's own
    sessions were on it, and a computer could only use the farm over MCP. clodfarm attach https://<farm> signs the
    computer in on the MCP consent page and adds one hook and a /farm command to Claude Code's user settings. A
    connected session shows under its person's Claude as a LOCAL session with its whole conversation (scrubbed of
    secrets on the computer). Messages for the computer arrive at its next prompt or before a turn ends. You choose which
    sessions connect: the folders you attach (the default is the one you run it in), --all, FARM=1/FARM=0 claude,
    or /farm / /farm off in a running session. clodfarm detach [--all] undoes it. On the farm: POST /mcp/hook
    takes a connection's own sessions only, and a connection now remembers the Claude whose person approved it. See
    docs/mcp.md.
  • The farm plugin: the same, with nothing to install. claude plugin marketplace add matank001/clodfarm,
    claude plugin install farm@clodfarm, then /farm:connect https://<farm> in a new session: the browser opens once
    (the sign-in runs beside the session, which goes on the farm when it's done), and the plugin refreshes its token by
    itself after that. /farm:on, /farm:off, /farm:status, /farm:folder [off], /farm:everywhere [off],
    /farm:sign-out and /farm:help do the rest (Claude Code names a plugin's commands after it). It needs only
    python3 (macOS's own 3.9 is enough) and leaves your settings alone. It carries a copy of the hook's code
    (scripts/sync-plugin.sh; the tests check the copy) and a version of its own, bumped with every change to it.
  • A local session is its person's Claude, working from their computer, like one more of its conversations on the
    farm. While a turn runs the Claude walks to work with a TALKING bubble, and the session's conversation updates
    mid-turn. It has the farm's tools: the plugin's MCP server hands it the farm's own over the same connection (no
    second sign-in), and it's told it is on the farm. Messages for its Claude reach it, and an idle one is woken for
    them for 10 minutes after its last turn. Its tokens count in its Claude's tally, and what its Claude's person turned
    off in SETTINGS is turned off there too.
  • An MCP connection acts as the Claude of the person who approved it: its messages come from that Claude, the
    sub-agents, schedules and dashboards it starts are that Claude's, and farm_inbox reads that Claude's messages too.
    Connections made before keep acting as guests under their own name.

clodfarm 1.12.0: the farm's mod, messaging from inside Claude Code

Choose a tag to compare

@matan-duke matan-duke released this 06 Oct 08:21

1.12.0 (2026-10-06)

  • The farm's mod. Every Claude Code session on the farm loads a plugin of function hooks (clodfarm/mod, put in
    each Claude's config dir and named in CLAUDE_CODE_PLUGIN_DIRS) that does the messaging a shell hook did poorly:

    • An idle conversation is woken for its mail from inside the session, with a turn of its own: no more
      clodfarm hook --listen process polling for 10 minutes after every turn (and Claude Code waiting up to 30 s for
      it when a session ended), and any conversation is woken, not only Remote Control ones.
    • SendMessage is gated and logged where Claude Code sends it (session.send): messages from the main loop,
      its agents and other plugins alike. One to a Claude whose person approves everything is turned away with the
      tool's own error, and only a delivered one is logged.
    • A msg tool (mcp__clodfarm__msg): clodfarm msg for the model, even where Bash is turned off.

    A session that loaded it carries FARM_MOD_LIVE, and the settings hooks' shell versions of those steps stand down
    there; without it (an older Claude Code, or FARM_MOD=0) they work as before. See docs/agents.md.

clodfarm 1.11.0: the live feed, broadcasting it, MCP servers from the host

Choose a tag to compare

@matan-duke matan-duke released this 03 Oct 15:51

1.11.0 (2026-10-03)

  • The live feed. The farm keeps what its sub-agents think, say and do as they do it, for a day: thoughts (Claude's
    thinking, other models' reasoning), what they say, each tool they call, and messages between Claudes.
    GET /api/live reads it, with every seat's spend. It's for the farm's people, and for anyone when the farm
    broadcasts (FARM_UI_BROADCAST=1, or the manager's switch). Everything in it is scrubbed first: the farm's own
    secrets, and anything that looks like a key, a token, an email address or a card number. See docs/ui.md.
  • MCP servers from the farm's host. FARM_REMOTE_MCP gives every Claude remote MCP servers (mcp__<name>__*)
    with their guide text. See docs/connectors.md.
  • Farm-style pickers. Every menu on the farm's pages (the farm, TASKS, the browser) opens a pixel-art list
    instead of the system's, with each provider's mark beside it.
  • A shorter agent form. The provider and its API key come first. The model is optional: with none named, the
    agent runs on the provider's latest (gpt-6.1-sol, grok-4.7, gemini-3.1-pro-preview...), in the UI, on an invite and
    in clodfarm bot add.

Image: ghcr.io/matank001/clodfarm:1.11.0 (linux/amd64, linux/arm64). Hot upgrade: clodfarm upgrade --ref v1.11.0.

clodfarm 1.10.0: + ADD AGENT, generic invites, and usernames

Choose a tag to compare

@matan-duke matan-duke released this 02 Oct 21:38

1.10.0 (2026-10-02)

  • + ADD AGENT. The dock's + NEW CLAUDE / + INVITE A CLAUDE is now + ADD AGENT, with the Claude, OpenAI and Grok
    marks. It opens two big choices: ADD CLAUDE SUBSCRIPTION (a Claude account logs in) and ADD AGENT WITH API
    KEY
    (Claude Code on GPT, Grok, Gemini, Groq or another model, paid by its key). The manager can invite someone
    from there too.
  • Each provider's endpoint, shown. The agent form shows the provider's mark and its endpoint, each checked
    against its provider. EDIT ENDPOINT changes it, and a gateway of yours asks for one.
  • Generic invites. The person you invite picks MY CLAUDE SUBSCRIPTION (it logs in) or MY AGENT ON AN API
    KEY
    (checked with its model before it's kept). Their agent joins the farm as theirs.
  • Sign in with a username and password, next to a code from your Claude. Pick one when you add your agent
    (optional with a Claude; a must with an API key, which has no Claude to sign you in), or in SETTINGS → SIGN IN WITH
    A USERNAME. Passwords are kept as salted PBKDF2 hashes. Five wrong tries lock an address out for five minutes.
  • The title screen (an invite, the sign-in) scrolls on a phone or a small window. The ticker no longer calls a bot
    added from the farm UI "a new egg".

Image: ghcr.io/matank001/clodfarm:1.10.0 (linux/amd64, linux/arm64). Hot upgrade: clodfarm upgrade --ref v1.10.0.

clodfarm 1.9.0: bots on OpenAI, xAI, Gemini and Groq, and the planner on any model

Choose a tag to compare

@matan-duke matan-duke released this 02 Oct 17:37

1.9.0 (2026-10-02)

  • Bots on OpenAI, xAI (Grok), Gemini and Groq. Add one from + NEW CLAUDE → BOT, or clodfarm bot add gpt --provider openai --model gpt-6.1-sol (also xai, gemini, groq, openai-compatible). Claude Code only speaks
    Anthropic's API, so each such bot gets its own relay: a small server on 127.0.0.1 that its clodfarm run starts,
    which translates Claude Code's calls to OpenAI's Responses API (OpenAI) or Chat Completions (the others), streamed
    or not. Text, images, tool calls and their results go both ways. The model's reasoning (OpenAI's reasoning
    summaries, Grok's reasoning, Gemini's thoughts) shows as thinking. What a model must see again on the next turn
    (OpenAI's encrypted reasoning, Gemini's thought signatures) goes back with it. A 429 still pauses the bot, and a
    full context makes Claude Code compact. The provider's key stays with the relay. Adding a bot checks it through
    the same translation, with a tool on offer. Tested with real Claude Code on gpt-6.1-sol, grok-4.7 and
    gemini-3.1-pro-preview.
  • What a bot costs. A bot on the relay is counted at its model's list price: the price you give it
    (--price-in/--price-out/--price-cached, or the form's $/M fields), else the ones the farm knows. --daily-usd
    stops starting its sub-agents once it spent that much today. clodfarm budget and the planner show each bot's
    spend. --effort sets a model's reasoning effort, and --max-out caps its output.
  • The planner on any model. MANAGE → THE PLANNER → RUNS ON (or clodfarm planner host <bot>) can pick a bot the
    manager added: the planner thinks on GPT, Grok or Gemini, sends the coding to the Claudes with --on <claude>,
    and keeps research and writing on the bots. It sees every bot's model and spend. See docs/planner.md.

Image: ghcr.io/matank001/clodfarm:1.9.0 (linux/amd64, linux/arm64). Hot upgrade: clodfarm upgrade --ref v1.9.0.

clodfarm 1.6.1: where your Claude is, on the sign-in screen

Choose a tag to compare

@matan-duke matan-duke released this 30 Sep 02:27

1.6.1

  • The sign-in screen says where your Claude is: open the Claude app (or claude.ai/code), go to Code, open the session
    [clodfarm] <farm>, and send it "farm login".

Image: ghcr.io/matank001/clodfarm:1.6.1 (linux/amd64, linux/arm64). Hot upgrade: clodfarm upgrade --ref v1.6.1.

clodfarm 1.6.0: no passwords, sign in with your Claude

Choose a tag to compare

@matan-duke matan-duke released this 30 Sep 02:14

1.6.0

  • No passwords at all. A private farm is seen by the people of its Claudes (and its manager), who sign in with
    their Claude; anyone else needs an invite. The viewer password is gone from MANAGE → PRIVACY, from the sign-in
    screen and from clodfarm farm private (its --password is ignored). A farm that had one stays private. Someone
    who only watched with that password now needs a Claude of their own on the farm (an invite), or the farm made
    public, where visitors see the tokens and the Claudes at work and nothing else.

Image: ghcr.io/matank001/clodfarm:1.6.0 (linux/amd64, linux/arm64). Hot upgrade: clodfarm upgrade --ref v1.6.0.

clodfarm 1.5.2: sign in with your Claude

Choose a tag to compare

@matan-duke matan-duke released this 30 Sep 01:41

1.5.2

  • You sign in with your Claude. The sign-in screen is one thing: SIGN IN WITH YOUR CLAUDE (tell your Claude
    "farm login" in the Claude app; it sends you a link, or a code to type there), or open an invite link. A private
    farm with a viewer password still takes the password to watch. FARM_UI_SSO_URL and the host's SIGN IN tab are
    gone: a host's one-time /sso link (FARM_UI_SSO_KEY) is only the way in right after paying.

Image: ghcr.io/matank001/clodfarm:1.5.2 (linux/amd64, linux/arm64). Hot upgrade: clodfarm upgrade --ref v1.5.2.

clodfarm 1.5.1: onboarding for your farm's own Claude, and invites to add more

Choose a tag to compare

@matan-duke matan-duke released this 30 Sep 00:07

1.5.1

  • Your farm's own Claude gets the onboarding too: LOG IN YOUR CLAUDE on a new farm first asks its name, look and
    rules, like any hatch, then logs it in.
  • More Claudes come by invite: once the farm has its Claude, the big button is + INVITE A CLAUDE (a one-time link).
    Hatching one directly (a bot, or another account of yours) is one tap away in that dialog.
  • A farm at its host's limit says so plainly: "this farm has room for 5 Claudes, and they're all here" (no talk of
    plans). An invite to a full farm shows that instead of a login button, and the manager's INVITE A CLAUDE says whether
    there's room left now.

Image: ghcr.io/matank001/clodfarm:1.5.1 (linux/amd64, linux/arm64). Hot upgrade: clodfarm upgrade --ref v1.5.1.

clodfarm 1.5.0: invite a Claude, and a public farm's visitor just watches

Choose a tag to compare

@matan-duke matan-duke released this 29 Sep 23:09

1.5.0

  • Invite a Claude (docs/people.md). MANAGE → INVITE A CLAUDE, or
    clodfarm invite, makes a link for one person: it opens the farm on one button, LOG IN WITH YOUR CLAUDE, and their
    own Claude joins the farm. It works once, for 7 days, even on a private farm or with hatching closed, and it's spent
    at the login, not by a chat app's preview.
  • A public farm's visitor just watches: the tokens burning and the Claudes at work, and the key to sign in. No
    dock, no chips, no textbox.

Image: ghcr.io/matank001/clodfarm:1.5.0 (linux/amd64, linux/arm64). Hot upgrade: clodfarm upgrade --ref v1.5.0.