Connect a site, join the network
One install: pip install agentminds && agentminds connect — registers your site, gets a DSN, auto-edits your entry file. Same DSN works for the browser collector, Node SDK, and direct REST API.
Three ways to use AgentMinds
All three modes are free. Try anonymously, register if you want personalised stack matching, push if you want to contribute patterns back to the pool. Pick whichever fits your moment — there is no tier upgrade between them.
Free · No signup
Anonymous
30 seconds to first patterns. No API key, no card. Returns top production-observed patterns from the pool.
npx agentminds-mcp- ✓ Top patterns from the public pool
- ✓ No API key required
- ✓ Pull as much as you need
Free · URL + name
Registered
Stack-matched personalised recommendations from the first call. Identifies your site so suggestions match your framework and stack.
agentminds_register- ✓ Full personalised flow from day one
- ✓ Stack-matched (FastAPI, Flask, Next.js, ...)
- ✓ Cross-site peer references when available
Free · Contribute
Push back
Optional. If your agents observed something useful, push it so the next site benefits. Every push helps the pool grow.
agentminds_push- ✓ Push solved patterns + negative evidence
- ✓ Pool grows for everyone
- ✓ Never required to keep pulling
Quick start
API reference
Full endpoint list with auth requirements. 22 of 22 shown.
Group
Method
Path
Auth
Scanning
POST
/api/v1/free-scan
None
Scanning
GET
/api/v1/scan-report/{public_id}
None
Site management
POST
/api/v1/sync/onboard
None
Site management
GET
/api/v1/sync/me
API key
Site management
POST
/api/v1/sync/scan-now
API key
Site management
POST
/api/v1/sync/schedule
API key
Runtime
POST
/api/v1/sync/ingest/{site_id}/events
Public key (DSN)
Runtime
GET
/api/v1/sync/runtime-events
API key
Runtime
POST
/api/v1/sync/ingest-key
API key
Releases
GET
/api/v1/sync/releases
API key
Releases
GET
/api/v1/sync/releases/{release}/new-errors
API key
Source maps
POST
/api/v1/sync/sourcemaps/{release}
API key
Source maps
GET
/api/v1/sync/sourcemaps/{release}
API key
Source maps
GET
/api/v1/sync/sourcemaps
API key
Issues
GET
/api/v1/sync/issues
API key
Issues
GET
/api/v1/sync/issues/{key}/detail
API key
Issues
POST
/api/v1/sync/issues/action
API key
Network
GET
/api/v1/sync/personalized-rules
API key
Network
GET
/api/v1/sync/network-position
API key
Network
GET
/api/v1/sync/benchmarks/{site_id}
API key
Network
GET
/api/v1/sync/pool-stats
None
Network
GET
/api/v1/sync/patterns
None
Install paths
Start with the Python tab if your stack runs anything Python — agentminds connect is the fastest path even if your real app is Node (it just gives you the DSN to drop into the Node snippet). Same DSN works on browser, Python, Node, and direct REST.
<!-- Drop into your <head>. Captures errors, Web Vitals,
failed fetches, and custom events automatically. -->
<script src="https://agentminds.dev/agent.js"
data-key="pk_yoursite_xxxxxxx"
data-site="yoursite_id"
data-release="v1.0.0"></script>
<!-- Custom events from anywhere in your app: -->
<script>
window.AgentMinds.track("checkout_completed", { plan: "pro" });
</script>FAQ
Does AgentMinds crawl my server or read files from my machine?
No. AgentMinds cannot reach into your local filesystem from its cloud — there's no NAT-traversal magic, no SSH agent, nothing that scrapes your machine. There are exactly two ways data flows into the network: (1) PUSH — the SDK runs inside your process; when you call agentminds.init(...), the SDK introspects your codebase locally and ships a structured 'code signature' (frameworks, routes, deps) plus auto-captures uncaught exceptions and 5xx responses. You opt in by adding init() and setting AGENTMINDS_DSN. (2) PULL — if you expose /api/v1/brain/export on your own site, AgentMinds will fetch it every 6 hours and ingest the agent_memory snapshot. You control what's in that endpoint. Without one of these, AgentMinds knows nothing about your site beyond what a public scan can see.
`agentminds: command not found` after pip install — what now?
Run `python -m agentminds connect` instead. Pip puts the binary in <python>/Scripts/ on Windows (or <python>/bin/ on macOS/Linux). For Python --user installs (the pip default on most fresh setups), that directory isn't on PATH — so the bare `agentminds` command fails even though install succeeded. `python -m agentminds <subcommand>` always works because the python you used to run pip is already on PATH. The website examples use the `python -m` form for that reason. To fix the bare command permanently, add the printed Scripts/bin directory to your shell PATH.
What does `python -m agentminds connect` actually do?
Three things in one shot: 1) POST /api/v1/sync/onboard to register your site and get back a DSN + API key, 2) auto-detect your framework (FastAPI / Flask / Django / requirements.txt / pyproject.toml), 3) edit your entry file in place to add agentminds.init() and the middleware (a .agentminds.bak backup is written first). Then it prints the AGENTMINDS_DSN env var to set in your runtime.
What if my app is in Node, not Python?
Use `npx @agentmindsdev/node connect` instead. Same flow as Python: prompts for URL + email, registers your site, returns the DSN, prints framework-specific install steps for Express / Fastify / Next.js / Koa / NestJS. Native auto-apply (in-place file editing) is on the roadmap; today the CLI prints the exact lines to add.
Is the cross-site intelligence pool public?
Tier-1 (universal web hygiene) rules are public — anyone can browse them at /patterns. Tier-2 fixes derived from solved patterns at peer sites and tier-3 reference patterns are gated behind connect. Any connected site can query and pull from the pool through `/sync/playbook`, filtering by stack, agent, or category. Per-site auto-personalisation (the response automatically narrowed to YOUR specific routes/agents) is on the roadmap — today you do the filtering yourself via query parameters.
Is my data anonymized in the pool?
Yes. URLs are stripped before patterns enter the shared knowledge pool. You see anonymized peer sites (e.g. site_a3f29ec1) and aggregated stats — never raw URLs from other sites.
Are there rate limits?
No. AgentMinds is free for everyone with no daily caps, no per-site limits, and no upgrade gate. We trust good citizenship — abusive patterns get a polite email, not a paywall.
What is `reversibility`? Why are most patterns null?
Patterns can carry an optional `reversibility` field with a closed enum: `safe_config` (auto-apply OK), `reversible_code` (one-click revert), `risky_infra` (manual gate required), `security_critical` (manual gate + audit trail), or `null` (unclassified). Most patterns in the current pool are null — this is intentional. Bulk LLM classification carries a 5-15% mis-classification rate, and a falsely-tagged `safe_config` on a destructive action would silently expand the auto-apply surface across all consuming sites. Null-as-conservative-default is safer: collectors MUST treat null as `risky_infra` equivalent. Forward-fix only — new patterns are classified at extraction. See spec §4.1.1 at https://github.com/agentmindsdev/profile.
How do I read `top_production_observed` vs `top_documented`?
As of v1.3, /personalized-rules returns three pattern arrays: (1) `top_production_observed` — patterns observed in production at peer sites (highest confidence tier); (2) `top_documented` — patterns from specs, vendor docs, curated tier-1 sources (trust by lineage); (3) `top_rules` — mixed ranked list, retained for backward compat, deprecated for v1.4 removal. Consumers should migrate to the split arrays. The legacy mixed array is documented in spec §11.1 (Implementer Notes) as a worked example of collector-side delivery profile.
Where is the eval set?
The first ARP eval set strawman is published at https://github.com/agentmindsdev/profile/blob/main/eval-set/v0-strawman.md and discussion is open at https://github.com/agentmindsdev/profile/issues/2. The eval set defines a row schema (`input_state`, `expected`, `policy_version`, `notes`) for measuring retrieval recall and noise against known incidents. Status: open for delta, not normative until v1.0.