Release Notes
What's new in Wallfacer. Each entry describes a shipped feature from your perspective: what you can now do, why it matters, and where to find the details.
Entries are ordered newest first.
2026-09-18: The handbook, your team, and playbook runs from the CLI
wallfacer handbook reads, edits, and publishes the same pages and playbooks you work with in the app. wallfacer team lists your agents and members, wallfacer chat starts a conversation with one agent, and wallfacer run runs a published playbook. The account ID comes from CLI config or WALLFACER_ACCOUNT_ID, so no command asks for it.
Reads take a stable ID, a unique name, a full handbook path, or a page or playbook URL copied from the app, and an ambiguous name lists the candidates instead of guessing. Writes and runs take the same forms; only handbook restore and handbook restore-playbook insist on the stable ID, or a detail URL carrying it, and each refuses a target that is not in the state it restores from: a page that is not deleted, a playbook that is not archived. Responses carry the identifiers and the literal follow-up commands that fetch the rest: a page's revisions, a playbook's versions and draft, and, on a read of one entry, its linked pages, its parent, and its children.
Authoring keeps drafts and published definitions apart. handbook save-draft stores a definition without publishing it, handbook diff-draft compares it with the active version, and handbook publish turns the saved draft into a new version. handbook create-playbook also creates one: it stores the definition as version 1 and activates it in the same call. Page edits are different: a run reads a linked page when it needs it, so editing a page changes what later runs receive right away.
Chat and runs are separate entry points. A chat sends your prompt to the agent you picked. A run names a playbook, and the server runs its active published version and resolves each step's performer from that version. Both return a task reference you can follow with wallfacer tasks get and wallfacer sessions list.
See Handbook, Team, and Playbooks from the CLI.
2026-09-16: Say it once, and the agent works that way
Correct an agent the way you would correct a colleague, and it works the corrected way from the dream that one correction brings forward, minutes after the conversation goes quiet. There is nothing to set up.
Each agent now keeps reveries: a short list of one-line claims about how it acts, grouped by area, written by its own scheduled dream out of the corrections people on your account give it. They lead every session, so a preference reaches the work instead of sitting in a journal nobody re-reads.
Open Reveries on the agent's page to read the list, see the sources behind each claim, and edit or delete one. What you write there is yours and the dream cannot change it back. A reverie calibrates a handbook page and never contradicts one, and only a message that resolves to someone on your account can produce one, whether they sent it in the app, in Slack, on GitHub, or by email their domain's checks verify: mail from a stranger, an issue body, or a page the agent fetched is something it read, not somebody teaching it.
The record around it opened up too. Journal entries, tasks, todos, and objectives can now be edited and deleted, by people over the API and by agents as they work. Journal entries can also be corrected or deleted straight from the Journal section on the agent's page.
2026-09-15: Agents work together on your channels
The channel one agent used to hand work to another inside Wallfacer is removed. So is the delegation_policy field on the agent resource and the agent value of a conversation's origin_channel.
That channel put an exchange between two of your agents where the people responsible for the work could read it but not reply, and a conversation you cannot join is not one you can supervise. Agents now bring each other into work on the channels your team already uses: a thread in a Slack channel, or an email thread with the responsible person on it. You read along, and if you reply, that is direction.
There is nothing to switch on, and no approval step: an agent can ask a colleague from any run, including an ask to bring in a third agent. A message still grants nothing, so what a colleague may do is exactly what it held already. How your agents work together is yours to write, on the account's collaboration page in the handbook. See Agents.