tick

name	tick
version	0.0.0
description	Hosted runtime for frame datasets. Four ops — curate, refresh, verify, discover — billed via x402/MPP in USDC. Wallet IS the identity.
homepage	https://tick.frames.ag
metadata	{"moltbot":{"category":"agent-runtime","api_base":"https://tick.frames.ag"},"frame":{"protocol":"0.2.0","write_tools":["add_entity_with_facts","set_facts","deprecate_fact","attach_evidence"],"events":["entity.created","fact.set","facts.set_many","fact.deprecated","evidence.attached","entity.removed","tool.invoked"]},"payment":{"protocols":["x402","x402v2","mpp"],"chains":["solana","base","tempo"],"settlement":"USDC"},"mcp":{"server":"npx -y @frames-ag/tick mcp","tools":["runtime.curate","runtime.refresh","runtime.verify","runtime.discover"]}}

tick

tick is the hosted runtime layer for frame datasets. Point it at a frame URL and it will read, fetch, verify, and write evidence-backed facts on your behalf — billed per call in USDC, wallet-signed, no API keys.

Four operations:

Op	What it does	Default budget (USDC)
curate	Full agent loop: read state, search sources, write facts with evidence	1.50
refresh	Re-fetch every fact's source.url; deprecate dead sources, attach redirects	0.30
verify	Read-only: re-fetch sources, report drift. No writes.	0.15
discover	Search-only: propose candidate entities for human review	0.50

Customer pays upto budget; settles actual consumed.

---

TL;DR — Quick Reference

Frame URL format: https://github.com/<user>/<repo>[/<frame_path>]

Just want the cheapest answer? Use verify — pure read, never mutates.

Want your dataset updated? Use refresh (fixes link rot) or curate (adds new entities).

Authentication: wallet signature on the inbound x402 challenge. No API keys to manage.

---

Two integration paths

Path 1 — MCP server (recommended for harnesses)

Add to .mcp.json:

{
  "mcpServers": {
    "tick": {
      "command": "npx",
      "args": ["-y", "@frames-ag/tick", "mcp"],
      "env": {
        "TICK_API_URL": "https://tick.frames.ag",
        "TICK_API_KEY": "<wallet-session-token>"
      }
    }
  }
}

Then the four tools appear in any MCP-aware harness (opencode, Claude Code, Codex CLI, Cursor):

• runtime.curate({ frame, budget? }) → events + run_id + settled cost
• runtime.refresh({ frame, budget? }) → events + drift report + run_id
• runtime.verify({ frame, budget? }) → drift report + run_id
• runtime.discover({ frame, budget? }) → candidate entities for review

Path 2 — Direct HTTP

curl -s -X POST "https://tick.frames.ag/run" \
  -H "Authorization: Bearer ${TICK_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "op": "verify",
    "frame": "https://github.com/microchipgnu/frames-examples/datasets/mcp-servers",
    "budget": "0.15"
  }'

Response shape (success):

{
  "run_id": "run_a1b2c3...",
  "op": "verify",
  "frame": "https://github.com/microchipgnu/frames-examples/datasets/mcp-servers",
  "settled": "0.087",
  "events": [],
  "tool_log": [
    {
      "seq": 0,
      "descriptor_id": "tick:paid-refetcher",
      "tool_id": "tick.paid.fetch",
      "cost": "0.01",
      "source_url": "https://api.github.com/repos/...",
      "retrieved_at": "2026-05-11T16:01:54Z"
    }
  ],
  "summary": "verify · mcp_servers@a1b2c3d · 13 entities, 47 fields checked · 36 verified, 11 drifts · $0.063 remaining",
  "started_at": "2026-05-11T16:01:00Z",
  "ended_at": "2026-05-11T16:01:54Z",
  "report": { "drifts": [...], "stats": {...} }
}

---

When to use which op

curate — full agent loop (paid LLM + paid tools)

Use when:

• New entities need to be discovered + added to the dataset
• A daily tick that does discovery + refresh in one shot
• You have a prompt.md describing the loop and want it executed end-to-end

Cost shape: 30+ LLM calls + 10+ paid tool calls is typical. Default $1.50 budget covers P50 cycles; bump for discovery-heavy runs.

refresh — re-verify and mutate

Use when:

• You want to detect link rot and emit fact.deprecated events
• You want to attach redirect-destination evidence to existing facts
• A cheap weekly tick to keep facts honest

No LLM cost. Tool cost varies with how many fields you have.

verify — read-only drift report

Use when:

• You want a drift report without writes
• Quick sanity check before deciding to refresh
• CI gate that says "this dataset is still well-sourced"

Cheapest op. Returns structured drifts (source_dead / value_drift / excerpt_missing / source_redirect).

discover — propose new entities

Use when:

• You want candidates surfaced but reviewed by a human before commit
• Doing a search-heavy initial seed for a fresh frame
• Bounded "find me 5 more" without trusting the agent to write directly

Returns candidates with evidence; writes go to a review queue, not the frame.

---

How payment works

Two modes, both settled in USDC, wallet-native:

• /run = single-shot x402 upto — verify+settle in one call. Best for verify/refresh/most curate ops.
• /session = multi-turn MPP — pre-authorize a cap, stream charges, settle on close. Best for long curate runs or stateful work.

Tick supports both protocols transparently. Solana + Tempo MPP work today (via payment-tempo). x402 works on Solana + Base. No Stripe dependency.

---

Receipts (the join story)

Every run produces a run_id. The receipt at GET /runs/<run_id> is publicly queryable by run_id possession. The receipt contains:

• All frame events written during the run (with run_id baked into each envelope per frame protocol v0.2.0)
• All paid tool calls with source URLs, costs, and timestamps
• Total settled amount

Anyone reading the customer's frame can join events.ndjson to our run log by run_id — see exactly which paid tool produced each source URL, at what cost, when. This is what makes "evidence-tracked" verifiable.

---

Failure handling

• Source dead during refresh: emit fact.deprecated, settle the part of the budget consumed
• Tool fails mid-curate: agent receives an error tool_result, may retry with a different tool
• Budget exhausted: agent gets a system message asking it to wrap up; loop force-stops
• Frame unreachable (frames-cloud can't resolve): run returns 4xx, no payment settled

upto semantics mean you never pay more than your declared budget. Partial completion settles for the partial amount.

---

Rate limits

Per-wallet rate limits applied at the gateway. SIWX-gated read endpoints (/runs/<id>, /history, /balance) are free up to per-wallet limits. Paid ops carry their own internal rate-limiting on the agent loop side (max iterations, max tool calls per turn).

---

Tips

• Bulk-write whenever atomic. If you can compute many fields from one source, use a single add_entity_with_facts or set_facts call. Emits a single facts.set_many event instead of N — cleaner audit log, fewer events.
• Cite real sources. The runtime rejects facts without source.url + source.retrieved_at. Don't invent URLs.
• Use excerpt liberally. A verbatim quote in source.excerpt is the single biggest signal-to-noise improvement for downstream verify runs.
• Migrate from opencode/Claude Code: swap your OPENROUTER_API_KEY env var for TICK_API_KEY and replace opencode run with npx @frames-ag/tick curate <frame-url>. Same prompt.md, same dataset, just paid via wallet instead of API key.

---

Related skills

• agentwallet — provision the wallet you use to authenticate
• registry — frames registry for paid tool discovery (tick uses this internally via the catalog)
• analysis — philosophical analysis framework used by blindspot.news (one of tick's canonical workloads)

---

Roadmap

• curate and discover ops (live as of 2026-05-11; classification quality improves week-on-week)
• Per-tier MCP scoping (each dataset declares its tool needs at boot)
• Multi-provider model routing (OpenAI, Gemini, open models via AI Gateway BYOK)
• Sandbox integration for code-running tools inside curate
• Per-session sub-wallets for stronger custody isolation

See PLAN.md for the full build state.