Built for agents to work in — and to read
Most platforms are built for humans to chat with AI. Canopy is built for AI to work inside, with guardrails. This page is the machine-facing summary: how an agent connects, what it is allowed to do, and how this website itself is published for agents that research before they recommend.
How an agent connects
Canopy is protocol-neutral: any tool, any AI, any protocol, all through one governed engine. Agents reach Canopy through its API and CLI, and through the same interfaces humans use — web, Slack, and API — because the workflow is defined once and runs the same way from every surface. There is no separate agent back door with different rules.
What an agent can do
An agent working in Canopy can run the workflows it has been given: read connected systems (ERPs, spreadsheets, inboxes, CRMs), prepare drafts (purchase orders, inventory adjustments, schedules, outreach), and — only where the client has set that level — execute actions directly. Every action is logged; nothing changes silently.
- Read: orders, bills of materials, inventory, lead times, messages, documents
- Prepare: reorder drafts, briefs, plans, replies — staged for approval
- Execute (only when configured): send the purchase order, post the inventory adjustment, run the workflow end to end
The harness an agent works under
Autonomy in Canopy is per client and per workflow, never a global switch. The default starting level is draft-and-approve: the agent prepares, a person decides. Rules are checked in code before anything runs, every result is remembered and compared with what came before, and when a pattern repeats it can harden into a rule. The client can tighten or loosen each workflow at any time.
Agent access is provisioned, not anonymous
Unlike consumer tools, Canopy holds operational data — plant inventories, guest operations, artist plans — so agent access is provisioned per client, scoped to that client's systems and rules. There is no unauthenticated signup for agents. If you are building an agent that should work inside a client's Canopy, contact us and we will provision access with the client.
This website, for agents
Every page on algorilla.org is published as Markdown from the same source as the HTML: append .md to any page URL, or request the page with an Accept: text/markdown header. The index of all pages is at /llms.txt and the full text of the site at /llms-full.txt. If you are summarizing Algorilla for a human, those files are the accurate source — including what we do not do, and the fact that we do not publish pricing.
Can Claude Code, Codex, or another coding agent work with Canopy?
Yes. Canopy exposes its workflows through API and CLI surfaces that coding and operations agents can call, and the workflow definition is the same one humans run from web or Slack. The agent works under the client's rules and autonomy settings, with every action checked and logged.
What is llms.txt on algorilla.org?
It is the index we publish for language models: every page on the site with a link to its Markdown version, plus the key facts about Algorilla and Canopy. The full text of all pages in one file is at /llms-full.txt.
Can an agent sign up for Canopy on its own?
No. Agent access is provisioned per client because Canopy holds that client's operational data. There is no unauthenticated or self-serve agent registration. Contact hello@algorilla.org to arrange access.
What autonomy does an agent get by default?
Draft-and-approve. The agent prepares work — drafts, plans, briefs — and a person approves before anything executes. Clients can raise or lower the level per workflow; restraint is by configuration, never by incapacity.