Pickaxe Learn
Everything we've shipped to the Pickaxe platform, newest first — what each change does, and why you'd reach for it.
Three new open-weight models joined the model selector, available to every agent in your workspace.
Why it matters. They land near the top of the quality charts at a fraction of the credit cost — worth testing on high-volume agents where a frontier model is overkill.
The invite modal now takes an access group, so a new member lands in the right tier the moment they accept.
Why it matters. Saves the two-step dance of inviting someone and then hunting for them in the members table to assign their plan.
YouTube transcripts added to a knowledge base can keep their timestamps.
Why it matters. Your agent can cite the moment in the video an answer came from, which turns a long recording into something a user can actually verify.
The CLI onboarding step now offers to hand the whole setup to Claude Code, Codex, or a plain terminal, instead of assuming you'll type the commands yourself.
Why it matters. If you already live in a coding agent, you can go from zero to a configured workspace without reading the install docs.
Monetization settings now carry a currency, so access group pricing and checkout render in the currency you choose.
Why it matters. Charging a European or UK audience in their own currency removes the mental conversion at checkout and cuts the drop-off that comes with it.
Documents live at the workspace level and get connected to whichever agents need them, instead of being uploaded per agent.
Why it matters. One upload of your handbook, pricing sheet, or policy doc now serves every agent — and updating it once updates it everywhere.
Connect Notion or SharePoint over OAuth and pull pages and documents straight into a knowledge base.
Why it matters. If your team's real documentation already lives in Notion or SharePoint, you can point agents at it instead of exporting files and watching them go stale.
The Add Files modal was rebuilt around source types: upload an entire folder at once, or add a YouTube video, an RSS feed, or a block of pasted text.
Why it matters. Most knowledge isn't a tidy PDF. Feeding an agent a whole folder or a live feed beats converting everything into files first.
Any agent can be deployed as an MCP server your customers connect to from Claude Code, Codex, Cursor, VS Code, and other MCP clients — with the setup instructions generated for them.
Why it matters. Your expertise stops being a website people visit and becomes a tool that lives inside the environment where your customers already work.
Add an agent to a Google Meet, Zoom, or Microsoft Teams call — by pasting a link or connecting Google Calendar — and it records the conversation, helps in chat, and writes up a report afterwards.
Why it matters. The post-meeting report lands with decisions, action items, owners, and due dates already separated out, and the transcript files itself into Documents or the agent's knowledge base.
The OpenRouter catalogue is no longer gated to the top tier — every paid plan can select those models.
Why it matters. Access to the long tail of specialist and open-weight models without having to upgrade first.
Anthropic's Opus 5 is selectable on any agent.
Why it matters. The model to reach for when an agent is doing multi-step reasoning and getting it wrong is expensive.
Plan changes happen in-app: upgrades are prorated immediately, downgrades are scheduled for the end of the billing period.
Why it matters. No support ticket to move up a tier for a busy month, and no surprise charge when you move back down.
Two new Gemini models joined the selector, and reasoning became available on Gemini 3.5 Flash and 3.1 Pro Preview.
Why it matters. Flash Lite is the cheapest way to run high-volume classification and routing; the reasoning toggle lets one model cover both quick replies and harder work.
The Knowledge Explorer shows the individual chunks a document was split into and which ones came back for a given query.
Why it matters. When an agent answers from the wrong passage, this is how you find out whether the problem is your document, your chunking, or your prompt.
OpenAI's GPT-5.6 family arrived in three variants, each selectable per agent.
Why it matters. Three points on the speed-versus-depth curve from one provider, so you can tune cost per agent without changing how your prompts behave.
Claude Fable came back to the model list alongside Claude 5 Sonnet.
Why it matters. Fable is the one to pick when tone and narrative voice matter more than raw reasoning; Sonnet is the everyday workhorse.
Microsoft joined the login methods your end users can authenticate with.
Why it matters. For corporate audiences on Microsoft 365, it removes the password step that quietly kills adoption inside big organisations.
The Action Runs list is searchable.
Why it matters. Debugging a failed integration used to mean scrolling. Now you go straight to the run you care about.
Problem users can be blocked by email address or by IP from the Users table.
Why it matters. The lever you need when someone is abusing a public agent and burning your credits doing it.
Tabs are grouped into four areas — Agents, Deployments, Users, and Settings — each holding the things that belong together, with a dedicated deployments page.
Why it matters. The old flat tab bar stopped scaling as the platform grew. Related work now lives in one place instead of three tabs apart.
A new access group type where nobody can join by signing up — members exist only because you invited them.
Why it matters. The right shape for a client workspace, a pilot cohort, or an internal team, where an open signup form would be a liability.
An access group can declare which groups it upgrades into, and which portals it grants access to, with the upgrade offer previewed as your users will see it.
Why it matters. Lets you build a real ladder — free tier to paid tier to premium portal — instead of a set of disconnected groups you wire together by hand.
The access group list gained type filters and sorting, with a table layout that shows member counts at a glance.
Why it matters. Once you're running more than a handful of tiers, finding the one you need shouldn't take a scroll.
Dozens of connectors — Google Drive, HubSpot, Notion, DocuSign, Slack and more — can be wired up straight from the Actions tab, authenticated with either workspace credentials or each end user's own login.
Why it matters. This is what turns an agent from something that answers questions into something that files the ticket, updates the CRM record, and sends the contract.
Anthropic's Fable 5 became available across the model selector.
Why it matters. Built for long-form writing and character work — the difference shows most on agents that produce prose a human will actually read end to end.
An SSO endpoint creates an embedded session for a user you've already authenticated, so an embedded agent knows who it's talking to.
Why it matters. Your app's logged-in users don't have to log in a second time, and the agent can safely personalise on identity you vouched for.
Agents can be snapshotted, published from a snapshot, and reverted to an earlier one.
Why it matters. Change a prompt, watch quality drop, put it back — without reconstructing last week's configuration from memory.
OpenAI's second-generation image model became available to agents that generate images.
Why it matters. Noticeably better at text inside images and at following a layout brief — the two things that made the previous generation hard to use commercially.
Form-type agents can show a button that resets every field at once.
Why it matters. For agents people run repeatedly — quote generators, brief builders — clearing by hand between runs is the friction users complain about first.
An embedded agent can read the structure of the page it's on and act on it — clicking buttons, links, and toggles, scrolling, refreshing, and running multi-step sequences from instructions in its prompt. Works on every embed type except iframe.
Why it matters. This is the difference between an agent that explains where the setting is and one that goes and changes it for the user.
A planning tool that projects twelve months of revenue, cost, margin, and profit for an agent, with a usage histogram and inputs you can push around.
Why it matters. Find out whether your pricing survives a heavy-usage cohort before you launch it, not after the first invoice.
Gemini 3.5 Flash arrived, tuned for workflows where an agent chains several tool calls together.
Why it matters. Fast and cheap enough to sit inside a loop, which is where the cost of an agentic workflow usually gets away from you.
Charge your users by what their usage actually costs — 10,000 credits to $1 of Pickaxe usage — and set your own rate above or below that, including selling credit top-ups directly.
Why it matters. Flat subscriptions break the moment one user does fifty times the work of another. Credits keep the heavy users profitable without punishing the light ones.
Two new deployment targets, each about a minute to set up from the deployments page. Same agent, same behaviour, new surface.
Why it matters. Communities that live in Discord and audiences that live in Telegram shouldn't have to come to your website to use your agent.
Message Insights splits response time into Pickaxe latency, model time-to-first-token, total time-to-first-token, and generation time.
Why it matters. One aggregate number tells you an agent is slow. These four tell you whether to trim the knowledge budget, switch models, or shorten the output.
xAI's Grok 4.3 became selectable in the builder.
Why it matters. Strong on current-events reasoning, which matters for agents that work with fresh information rather than a fixed knowledge base.
Mistral's Medium 3.5 joined the model list.
Why it matters. A solid European option for teams with a preference — or a requirement — about where their model provider is based.
Deployments can fire webhooks on conversation events, so another system hears about them as they happen.
Why it matters. Push a completed conversation into your CRM, your data warehouse, or a Make or n8n scenario without polling the API for changes.
A copilot that takes real actions in your workspace — building an agent end to end, styling an embed, creating pages, managing access, updating knowledge bases — with no terminal, API keys, or setup.
Why it matters. It picks up your workspace credentials and drives the CLI for you, so the things that used to need a developer are a sentence instead.
Public and member access groups can be given a time-limited trial from the Setup tab.
Why it matters. An expiring trial creates a deadline. People who'd happily sit on a free tier forever tend to decide once there's a date attached.
OpenAI's GPT-5.5 became available across the builder.
Why it matters. The general-purpose default for agents where you'd rather not think about model choice at all.
Two smaller GPT-5.4 variants joined the selector.
Why it matters. For classification, routing, and short replies, Nano does the job at a fraction of the credits — and the quality gap only shows on hard work.
The CLI gained an action creation wizard, action manifests, connect-variable flags, and commands for MCP servers — plus custom page and portal commands the following week.
Why it matters. Integration setup becomes something you can check into a repo and replay, rather than a sequence of clicks you have to remember.
Programmatic access at both the agent and workspace level — fetch answers, add users, write memories, upload documents — opened up beyond Pro. Gold, discount, and trial accounts pay 1¢ per request; Pro has no per-request cost.
Why it matters. This is the hook for Make, n8n, and anything else you run: your agents become a step in a larger workflow instead of a destination.
Anthropic's Opus 4.7 landed in the model list.
Why it matters. The strongest option at the time for agents doing long, multi-step reasoning over your knowledge base.
A command line interface for building and updating everything in a workspace — agents, documents, deployments, custom styling — driven either by you or by a coding agent like Claude Code.
Why it matters. "Make me an agent that helps me sell real estate" becomes a real workflow, and custom portal styling stops being a design-tool problem.
xAI's reasoning model became available inside the builder.
Why it matters. Worth a look on agents that have to show their work rather than just produce an answer.
Embeds can carry injected custom code, configurable from the deployment or from the CLI.
Why it matters. Match your host site exactly, or run your own analytics and event tracking around the agent, without forking the embed.
API documentation was unified across the Documentation section and the deployment screens, with a completions tab and an Ask AI action on the docs themselves.
Why it matters. The reference you need is next to the deployment you're configuring, and you can ask it a question instead of reading the whole page.
Lock a single agent and sell access to just that, or lock a whole folder and sell the bundle. Locked content stays visible as a preview with an upgrade or unlock offer attached.
Why it matters. You can monetise a portal without restructuring it — and test whether people will pay for one specific thing before rebuilding your pricing around it.
A workspace-level toggle that automatically retries a failed request on a backup model. On for every plan.
Why it matters. Provider outages stop being your outage. The user gets an answer instead of an error, and you find out later.
Access groups can be priced as free alongside the paid intervals, and the pricing and welcome popups render it properly.
Why it matters. Lets a free tier sit in the same upgrade ladder as your paid ones, rather than living outside your pricing as a special case.
Select multiple agents — or all of them — and switch their AI model in a single operation.
Why it matters. Three real uses: move off a provider that's down, roll out a new model to test it broadly, or cut costs across the board in seconds instead of an afternoon.
Copy an entire workspace, and choose which access group new signups land in on the duplicate.
Why it matters. The core move for agencies: build a client setup once, then clone it per client instead of rebuilding it each time.
Google's Gemini 3.1 Flash Preview joined the model selector.
Why it matters. Very fast and very cheap — the model to try first when an agent's volume is high and each answer is short.
The end-user profile gained full plan management: downgrade, cancel with a save offer, restore a cancelled plan, and move between subscription and one-time purchases — on mobile as well as desktop.
Why it matters. Self-serve billing is what stops plan changes from arriving in your inbox, and the restore path recovers people who cancelled by accident.
An API endpoint for access groups, documented alongside the rest of the reference.
Why it matters. Provision and move users between tiers from your own systems — the piece you need when signup happens somewhere other than Pickaxe.
Tag users and access groups, then use those tags to organise and target who sees what.
Why it matters. Cohorts — beta testers, an enterprise client, a launch group — stop needing a separate access group each.
Build custom pages from editable sections — including agent and product listing sections — with a live preview and theme controls.
Why it matters. Your portal gets a real landing page, a pricing page, and a home page, without leaving Pickaxe for a website builder.
Copy a portal, and see a generated preview image of each one in the portals list.
Why it matters. Variants for different audiences start from your finished work instead of a blank portal.
Two models joined the allowlist: OpenAI's Codex variant and Google's Gemini 3.1 Pro Preview.
Why it matters. Codex for agents that read or write code; Gemini 3.1 Pro for long-context work over big document sets.
Deployments can be renamed, and the deployments list was rebuilt with clearer columns and empty states.
Why it matters. Once an agent has four deployments, "Direct Link" three times over stops being a useful label.
An extra-uses checkout flow lets end users top up when they hit their limit, presented as a carousel of options you configure.
Why it matters. Hitting a limit becomes a purchase instead of a dead end — and the users who hit limits are the ones who value the agent most.
Portals gained dark theme support and a dynamic glass treatment that adapts to the background behind it.
Why it matters. A portal that matches your brand's dark palette stops looking like a third-party tool bolted onto your site.
Sharing a deployment moved into a dedicated modal, with public and member access toggles alongside it.
Why it matters. Getting a link out to someone — and deciding who's allowed to open it — now happens in one place.
End users can sign in with an emailed magic link, configurable as a login method on the portal.
Why it matters. No password to forget is the lowest-friction sign-in there is, and it works for audiences that won't create an account.
Showing 67 of 67 updates. Releases are announced first in the Releases & Features category of the community.