← Dasha Compute · Run it yourself in 10 minutes

Proof, not promises.

This page has no marketing numbers. Every figure below is fetched live from a public endpoint when you load it, and each section links the exact endpoint it reads. Machine-readable twin: /compute/proof.json.

1. Right now

Source: /compute/api/network · /compute/api/readyz

community providers online

2. Every job, signed

Sources: /compute/api/verify · /compute/api/chain · /keys.json · format: /compute/llms.txt

receipts on the public ed25519 chain · verdict:
tip
receiptjobtokenscentswhen (UTC)

Recompute a receipt hash yourself - copy, paste, run:

curl -s https://www.getdasha.com/compute/api/chain | python3 -c '
import json,sys,hashlib
r=json.load(sys.stdin)["receipts"][-1]
body={"job_id":r["job_id"],"engine":r["engine"],"tokens":r["tokens"],"cents":r["cents"],"at":r["at"],"prev_hash":r["prev_hash"]}
h=hashlib.sha256(json.dumps(body,separators=(",",":")).encode()).hexdigest()
print("recomputed:",h); print("chain has :",r["hash"]); print("match:",h==r["hash"])'

Signature check: sig is ed25519 over the UTF-8 bytes of that hex string, verifiable against the signer key at /keys.json. Preimage format: /compute/llms.txt.

3. Ask the network

No signup, no key: the chain verdict is a public JSON endpoint. This is what it answered when you loaded the page:

curl -s https://www.getdasha.com/compute/api/verify

4. Fail loud, not silent

When no community Mac advertises the requested model, the Community v1 route fails loud with a machine-readable no_mac_online reason. The browser Hosted door is separate and may remain available:

{
  "status": "action_required",
  "reason": "no_mac_online",
  "hint": "Join a Mac or poll GET /compute/api/network.",
  "next": [ { "path": "/compute/api/network" }, { "path": "/compute#provide" } ]
}

Uptime history, including the dark windows - hourly presence samples from /compute/api/metrics (providers_online_hourly). Red means zero Macs online that hour:

5. What it costs

Source: /compute/api/pricing (live) vs DeepSeek's published pricing (checked Sep 12, 2026).

flat per successful chat completion, prepaid guest credits.
DeepSeek (per 1M tokens)off-peakpeak
deepseek-flash input (cache miss)$0.15$0.30
deepseek-flash output$0.60$1.20
deepseek-v4-pro input (cache miss)$0.66$1.32
deepseek-v4-pro output$1.98$3.96

From 12:00 Beijing on Sep 14, 2026, DeepSeek routes deepseek-v4-pro requests to V4.1-Flash and bills them at Flash rates - their table, their words, linked above. Different shapes: DeepSeek bills per token, Dasha bills a flat per-job price you can verify per job on the chain in section 2. DeepSeek is cheaper at scale; no one there hands you a signed receipt per job.

6. The provider wall, honestly

Source: /compute/api/network

Community side runs on spare Macs people plug in. Provider terms as published on the Provide door: $0.05/job + $0.01/1k completion tokens, $1 minimum payout, pending operator settle, +5% if you take $dasha. One more always-on Mac changes the reliability story more than any ad - become a founding provider.

Page rendered from live endpoints at . If an endpoint is unreachable its section says so instead of guessing.