Data & API

Everything on this site is built from one open dataset of 1005 systems. It's available as static JSON and CSV files: no key, no rate limit, licensed CC-BY-4.0 (credit "AI Agent Infrastructure Landscape" with a link).

Download a catalog

Catalog files are flattened to one row per system: name, section, tier, lifecycle status, licence, latest release, deployment, pricing, MCP/A2A support, the seven taxonomy axes and a link to the system page. For every cell with its citation and tier, use the per-system JSON below or the canonical landscape.json.

One system as JSON

GET ./api/systems/<id>.json

The id is the last part of a system page's URL. Example: Mem0. Each file has every cell (value, citation, status, tier), the taxonomy, incoming and outgoing edges, the change history and the lifecycle status the stack checker uses. api/index.json lists every id.

curl -s https://mrpeppersdev.github.io/agent-infrastructure-landscape/api/systems/mem0--mem0-ai.json | jq '.cells.license'

Change feeds

• All changes (Atom): the latest 100 of 935 recorded lifecycle, licence, release, pricing and hosting changes.
• One system: ./feeds/<id>.xml, also linked from each system page's "What changed" section.
• Your whole stack: run Check stack and choose "Watch these systems" to download an OPML file of feeds for every catalogued dependency.
• Recently verified (Atom).

From an agent or a script

• MCP server: choose_memory_layer, check_stack, find_alternatives, search_records, compare, recent_changes and more, for Claude, Cursor or any MCP client.
• CLI: landscape check package.json --fail-on-attention for CI, plus choose, alternatives and search with --json/--csv.
• llms.txt: a map of the site and data for LLM crawlers.
• Stack-checker index: npm, PyPI and Docker package names and GitHub repos mapped to systems and their lifecycle status.

How complete each column is

The share of systems with a value in each column. Many columns only apply to one kind of system (memory semantics to memory layers, citations to papers), so a low rate is often expected. Columns under 15% are hidden from the table until you ask for them; columns at 0% are planned and are left out of the exports until they have data. Also as JSON: /api/fill-rates.json.

Fixing the data

Every system page has a Suggest a correction link. It opens a pre-filled issue; a bot checks the source, patches the cell and opens a pull request for review. New systems go through Submit.