Ditto organizes the context your AI needs and operationalizes it across every thread, tool, and workflow.
SN 118. Rank 24 of 32 by emission
Ditto
Emission1.2%of the block reward · rank 24 of 32
Participants5 miners · 11 validators
Registration0.0005 TAO
Verified2026-09-09links checked and scores merged
The programagents · github.com@2b05b32 · read 2026-09-19
The work
A memory harness: the software that decides what an AI assistant stores, finds again and which tools it calls, with the language model held fixed. The miner packages it as a Docker build context serving a fixed HTTP contract and submits it for a platform-set fee, 0.1 TAO in the guide at this commit. Validators run it; the miner needs no hardware.
own docs · raw.githubusercontent.com
Miners submit a Docker build context that serves the public harness HTTP contract; the implementation may use Rust, Python, TypeScript, Go, or any other language.
How it is scored
Validators run the harness in a sandbox against a dataset regenerated from a seed every run: tool-calling cases and a memory haystack with distractors. Every harness is driven by the same locked model. Grading is by rule, no model as judge: composite = 0.5 tool accuracy + 0.5 memory recall, times bounded efficiency, consistency and canary gates. Finalized on the median across assigned validators; 100 eligible cases per entry minimum.
own docs · heyditto.ai
composite = (0.5 × tool_mean + 0.5 × memory_mean) × efficiency × consistency × canary
Sources disagree The site scoring page, updated 2026-07-13, says the locked model is Qwen3-32B; docs/MINER.md at the pinned commit says the current contract, Bench v12, is openai/gpt-oss-20b served through the platform's OpenRouter boundary.
How the pay splits
King of the hill with a tail. The champion takes 65 percent of the miner weight, the next four distinct miners 14, 10, 7 and 4, ordered by first-seen on ties. A challenger dethrones only by beating the champion by more than the larger of 0.007 composite points and 1.64 combined standard errors, on a median of 3 seeds. With no eligible miner, everything goes to the burn hotkey.
code · raw.githubusercontent.com · ditto/validator/config.py
KOTH_RANK_SHARES = (0.65, 0.14, 0.10, 0.07, 0.04); KOTH_MARGIN = 0.007; KOTH_DETHRONE_Z = 1.64
Sources disagree The site scoring page dated 2026-07-13 says the champion holds about 90 percent of emissions and the dethroning margin is about 5 percent; the repository at the pinned commit sets 65/14/10/7/4 and a 0.007 gate, and docs/MINER.md matches the code.
Who judges
Eleven validators were active on the 2026-09-09 chain snapshot. Commit-reveal is on. The platform's ledger assigns each submission to validators, who pull the screened image, run it in a sandbox and post signed scores; the ledger finalizes on the median and every validator recomputes the same weight vector from it with the same function. Weights go on chain through Pylon, an identity-based weight-setting service.
chain · hyperparams-2026-09-19.json
commit_reveal_weights_enabled True, weights_rate_limit 100, max_validators 64; chain-2026-09-09.json active_validators 11
How often it turns
Emission settles every 360 blocks, about 72 minutes. The validator sweeps the ledger every 30 seconds and pushes weights every 3,600 seconds. The puzzle turns per submission, not per epoch: a crown stands until a challenger clears the gate. A new registration is immune from deregistration for only 300 blocks, one hour, against the 5,000 most subnets use.
chain · hyperparams-2026-09-19.json
tempo 360, immunity_period 300; config.py VALIDATOR_EPOCH_SECONDS default 3600, VALIDATOR_SWEEP_SECONDS default 30
Who buys
End users of the Ditto app, paying per message in credits after 30 free messages, with no price table. The numbers page dated 2026-09-06 gives 2,505 registered users, 267 monthly active and an annual run rate of $6,648. Whether the winning harness runs inside the product is not stated on the pages fetched. The miner's income is emission.
own docs · heyditto.ai
Monthly active 267 last 30 days ARR $6,648 current run rate
What has been gamed
Memorizing the benchmark: the dataset is rebuilt from a seed every run, with the bench version mixed in. Self-reporting tool calls instead of hitting the grader's mock server: capped at 0.5 on those cases. Echoing a planted bait value: composite times 0.5. Copying the champion: first-seen timestamps plus near-duplicate detection, copies earn nothing. Dethroning on seed luck: the z-band and a 3-seed median.
own docs · heyditto.ai
a harness that skips the server and self-reports its calls is capped at 0.5 on the affected cases
3 of 7 fields rest on the chain or the code. A manifest is the site's reading of the subnet's program, not a review; each field links its source. All programs · What a field means · Correct a field
What it says
The subnet in its own words: the line the index took from its front door, and what the team has said on the record.
What it shows
Explained from zero by the index, from the subnet's own materials, for someone who has never heard of Bittensor.
SN118 Ditto runs a hiring contest for AI assistants' filing clerk: entrants submit software that decides what an assistant remembers and finds again, and the top entry takes most of the prize until something beats it.
The commodity, explained from zero
Two things are on offer, and they should be kept apart. The first is Ditto the product, from Omni Aura LLC: a memory service that gives "your AI a shared place to remember, retrieve, and act", which assistants such as Claude or Cursor plug into through an open connector standard called MCP. Its numbers page, updated 2026-09-06, reports 2,505 registered users, 267 monthly actives and $6,648 of annual recurring revenue. The first 30 messages are free, then customers pay per message in credits; the site's llms.txt still lists $5, $10 and $20 monthly plans that the pricing page has dropped.
The second is what the subnet produces: a memory harness, the clerk in the comparison. It sits between an assistant and its memory store and decides what to save, how to search and what to hand back. The README calls SN118 "A Bittensor subnet that incentivizes agent memory harnesses". Nobody buys a harness; the materials do not say whether the winning harness is used inside the Ditto product.
Why it is on Bittensor at all
The site's stated difference is "One unified memory across all AI assistants", a claim about the product against per-tool memory, not about decentralization. Nothing on the site says what the subnet adds. Here the comparison breaks: a hiring contest ends with a hire, while the Ditto champion is never hired, only paid until dethroned.
How the work gets done
Miners, the applicants, package a harness as a container image and submit it for about 0.04 TAO. Validators, the interview panel, run each harness in a sandbox against a freshly generated exam of tool-use cases and memory questions, every harness driven by the same locked language model, so the contest is about the clerk, not the brain behind it. Grading is by rule, with no model as judge, and the score is the median across validators. The best entry holds the crown, and a challenger takes it only by clearing a margin above measurement noise. Emission, the subnet's share of newly minted TAO, follows that ranking; with no eligible miner it is burned.
How you would know it works
DittoBench, dated June 2026 on the subnet's own site, tables ten commercial models on 177 tool-calling cases and 150 memory questions. The numbers page gives dated product figures. No public leaderboard of miner harnesses is readable without signing in.
What is missing
The two rulebooks disagree: the site's scoring page (dated 2026-07-13) says the locked model is Qwen3-32B, the champion holds about 90% of emission and the margin is about 5%, while the repository's miner guide says the model is gpt-oss-20b, the split is 65/14/10/7/4 across the top five and the margin is 0.007. The site page still says mining "is going live" in July 2026. The chain description, "Open-Source Claude Cowork", appears nowhere on the site. The chain contact peyton@omniaura.ai does not match the site's support@heyditto.ai. The chain's repository field points at a list of over twenty repositories, not the subnet code. There is no status page.
Go deeper
Sources for this explainer
- https://heyditto.ai/
- https://heyditto.ai/dashboard/
- https://heyditto.ai/pricing/
- https://heyditto.ai/llms.txt
- https://heyditto.ai/dittobench/
- https://heyditto.ai/docs/mining-on-ditto
- https://heyditto.ai/docs/mining-getting-started
- https://heyditto.ai/docs/mining-scoring
- https://raw.githubusercontent.com/ditto-assistant/ditto-subnet/main/README.md
- https://raw.githubusercontent.com/ditto-assistant/ditto-subnet/main/docs/MINER.md
Metaphor: a hiring contest for a filing clerk. Every claim is drawn from the evidence set or the subnet's own materials; "(inferred)" marks a conclusion rather than a quote. Corrections.
How legible it is
Four audiences, four questions each, rated on what a first-time reader can find in five minutes. The scale measures findability, not quality.
On-chain identity 7/7- subnet_name
- Ditto
- github_repo
- github.com
- subnet_contact
- peyton@omniaura.ai
- subnet_url
- heyditto.ai
- discord
- discord.gg
- description
- Open-Source Claude Cowork
- additional
- The Ditto Agent OS and Harness enable SOTA memory, healing, and speed for 1/100th of the frontier cost.
Stakers and validators●●●●● 3.8
Should I allocate here? · rank 6 of 32 for this audience
Q1What it is●●●●● 3
A Bittensor subnet that incentivizes agent memory harnesses.
output is the best harness, no unit or price, chain description Open-Source Claude Cowork is nowhere on the site, github_repo is an org
Q2Who it is for●●●●● 5
Ditto by the numbers.
Registered users 2,505, Monthly active 267, ARR $6,648, updated Sep 6, 2026, one click from the front page, consistent with the 2,500+ users on the front page
Q3How it resists gaming or fails●●●●● 4
Scoring and Rewards page: dataset regenerated every run, one frozen model, judge-free grading, median across validators, king-of-the-hill emissions with a hysteresis gate, no failure modes stated
Q4Identity and documentation●●●●● 3
identity 7 of 7, github_repo is the org repository list (resolves, the subnet repo must be found among 20 plus), url and discord resolve, description does not match the site pitch, no staker page
Miners●●●●● 4.0
Can I compete, and what wins? · rank 3 of 32 for this audience
Q1What it is●●●●● 4
You submit the whole crate.
memory tools contract with a starter kit as worked example, one click from the docs root, docs say a Rust crate, the subnet README says a Docker context in any language
Q2Who it is for●●●●● 4
The submission fee is a small transfer
of about $5 in TAO, the repo guide says 0.04 TAO plus a live registration preview (0.0005 TAO), no hardware needed (validators run it), a 0.007 gate to win
Q3How it resists gaming or fails●●●●● 4
Grading is fully deterministic. There is no language-model judge
tool and memory means weighted equally, integrity gates, median of validators, 0.007 hysteresis to dethrone, no examples, no churn
Q4Identity and documentation●●●●● 4
miner guide with prepare, submit, track and scoring sections, repo pushed 2026-09-10, docs pages dated July 13, 2026 still say mining goes live in July 2026, linked starter kit archived, no releases
Buyers and enterprises●●●●● 3.5
Can I use this today? · rank 10 of 32 for this audience
Q1What it is●●●●● 5
you can connect to Ditto's memory system right now using the Model Context Protocol (MCP)
endpoint https://api.heyditto.ai/mcp with a config example, app at assistant.heyditto.ai, mobile apps
Q2Who it is for●●●●● 3
Your first 30 messages are on us, then you pay only for what you use.
per-message credits with no price table, llms.txt still lists Flex $5, Plus $10 and Pro $20 plans the pricing page dropped
Q3How it resists gaming or fails●●●●● 3
Contact: support@heyditto.ai in llms.txt and a Support link in the footer, Discord on chain, no status page, no SLA, no response expectation
Q4Identity and documentation●●●●● 3
docs cover MCP server setup and a Developer Platform section, no REST reference beyond the MCP endpoint, support@heyditto.ai on the site does not match the chain subnet_contact peyton@omniaura.ai
Newcomers●●●●● 3.3
What is this and why does it matter? · rank 20 of 32 for this audience
Q1What it is●●●●● 4
Ditto gives your AI a shared place to remember, retrieve, and act.
on the front page, plain, but the chain description Open-Source Claude Cowork is a different pitch and the subnet is not mentioned
Q2Who it is for●●●●● 2
One unified memory across all AI assistants.
is the stated difference against per-tool memory, not a decentralization claim, nothing on the site says what the subnet adds beyond DittoBench
Q3How it resists gaming or fails●●●●● 4
DittoBench dated June 2026 with a table of 10 models on tool calling and memory on the subnet's own site, plus the Numbers page dated Sep 6, 2026, no live demo without sign in
Q4Identity and documentation●●●●● 3
subnet_name Ditto matches the site, url resolves, docs and blog serve as learn pages, but the chain description Open-Source Claude Cowork does not match the site pitch and there is no about page
Provenance
How this rating came to be. Verified means the scorer fetched the page and the words are on it; secondhand means concluded from code, absence, or a third party. The link badge is a separate automated check made before publication.
scored by agent d 2026-09-09, rubric v1.0merged 2026-09-09links verified 2026-09-09: 35 live, 1 unreachable, 2 manual
Something wrong? Corrections of fact are applied as they arrive during the window; rating disputes are batched at its close. How to file one · GitHub issue · email.