Agents published

4,812

▲ 6.2% this week

Seams in the index

1,374

▲ 46 this wk

Forks merged back

1,092

▲ 38 this wk

Runtimes surfaced to

▲ 2 this quarter

Spec v0.4 · open & stable

A repo is an agent. Layers, not lock-in.

STRATA is a git-native standard for defining AI agents. Identity, bounds, skills and memory live as plain files in a repository — versioned, diffable, reviewable. Fork an agent like you'd fork any library, then surface it to whatever runtime you actually use.

Official token AtJ351dUKdWz8MP1pQCK5UxyVqdkjyr61vXCgegQpump

Author rewards · this epoch

2,400,000 STRATA

Epoch settles in

hrs

min

sec

Forks today

1,318

Top author

vega.sol

Publishes · last 12 days

Layer stack

Core sample

Reading · vega.sol/atlas-researcher @ v2.4.0 Two layers required · eight optional, resolved top-down

strata.yamlrequired

CORE.mdrequired

BOUNDS.mdpresent

seams/6 seams

rigs/present

veins/present

deposits/absent

sediment/writing

assays/signed

beds/absent

Spec coverage8 / 10 layers declared

Fork drift vs upstreamCORE.md and 2 seams diverge

Index intelligence

Seam reuse across the registry

Every seam in the index, mapped by how many published agents import it. Hot cells are the load-bearing ones — the skills the ecosystem has quietly standardised on. Hover any cell for its import count.

rareload-bearing

How the standard works

Four moves, ten layers

Init

strata init lays down a manifest and a CORE.md. That repo is now an agent.

Layer

Add only what you need — seams, rigs, veins. Eight of the ten layers are optional.

Probe

strata probe validates the stack and refuses to sign a repo that contradicts its own bounds.

Surface

strata surface --to claude emits the native format for whatever runtime you're on.

Token board

Top 5 holders

Activity

Registry events

Last 7 days

Most-forked authors

#	Author	Forks	Seams published	Trend
1	vega.sol	312	24	▲ 3
2	7Kdq…mPXa	271	18	▲ 1
3	renji.sol	240	15	▼ 2
4	0xQuarry	190	11	▲ 5
5	4Rvn…t2Wc	172	9	▼ 1

Token

Where STRATA comes from, where it goes, and what ships next. Publishing and forking route fees straight back to the people who wrote the layers — the coin pays authors, it isn't a side bet.

The will

What STRATA stands for

An agent you can't read is an agent you can't trust.

No insiders

No VC rounds, no private sale. 30% floats free from day one and the team's 5% is locked for a full year. Nobody gets a head start on the index.

The coin works

STRATA isn't a mascot — it's how seam authors get paid. Every publish and fork routes a fee back to the people whose layers got used. If the index is alive, the coin is moving.

Supply only shrinks

A quarter of every protocol fee burns forever. 38.2M is already gone, and there is no mint authority — 1B is all there will ever be.

Built to be handed over

The roadmap ends with the keys given away: a DAO-curated index, an on-chain treasury, and holders ratifying the spec itself. Community is the exit plan.

Total supply

1.0B

fixed · no mint authority

Circulating

412M

▲ 41.2%

Burned to date

38.2M

▲ 3.82% of supply

Author rewards pool

214M

▼ paying out

Supply split

Tokenomics

Protocol economy

The flywheel

Publishing an agent or forking one costs STRATA. The split is automatic: most of it pays the authors whose seams got imported, a quarter burns, the rest funds the build. More reuse → better-paid authors → more seams worth reusing.

Supply math

The burn only compounds

Index pace 5 publishes / week

Solid line: 38.2M already burned across 4,812 publishes. Dashed: projection at the pace you set — every publish destroys ≈280k STRATA, permanently. Drag the slider, steepen the curve.

Live economy

Burn log

every fee splits in real time 25% burn · 60% authors · 15% build

What ships next

Roadmap

Q1 2026 · shipped

Spec v0.1 — two files

The manifest and CORE.md land, along with the reference CLI and the first surface adapter. 4,812 agents published since. Author rewards seeded with 30% of supply, LP locked at launch.

Q2 2026 · shipped

Seams & the import graph

Skills became first-class: versioned, imported by reference, and paid per import. Pinning went mandatory for releases the same month — unpinned imports now fail probe --strict instead of quietly drifting.

Q3 2026 · live now

Spec v0.4 — bounds & assays

Bounds are enforced at surface time rather than suggested, every surface writes a signed assay, and the loss report tells you exactly what an adapter dropped. You're looking at it.

Q4 2026 · next

On-chain provenance

Fork lineage and assay signatures commit on-chain, so an agent's ancestry is verifiable without trusting the registry. Author rewards settle automatically from the import graph — no human approves payouts.

Q1 2027 · planned

Beds & the resolver

Nested sub-agents get a real dependency resolver: conflict detection across imported bounds, deduped seams, and a lockfile. Plus typed clients and a local strata run sandbox.

Later · researching

A DAO-ratified spec

Holders stake STRATA to propose and ratify spec changes, and to curate what the index promotes. The standard stops belonging to us. Treasury governance moves on-chain with it.

Spec

STRATA v0.4. Two required files turn a git repository into a portable agent; eight optional layers add capability, knowledge and governance on top. Everything is plain text, so git diff tells you exactly what changed about an agent's mind.

Layout

Layers resolve top-down. Anything you don't need, you simply don't create.

# required — without these it is a repo, not an agent
strata.yaml            # manifest: model, version, deps, entry points
CORE.md                # identity: voice, values, how it communicates

# optional — add only what the agent actually needs
BOUNDS.md              # hard constraints, enforced at surface time
seams/                 # skills — loaded on relevance, not always-on
rigs/                  # tool schemas (MCP-compatible)
veins/                 # deterministic multi-step routes
deposits/              # reference knowledge
sediment/              # memory that survives the session
assays/                # signed audit record
beds/                  # nested sub-agents, imported by reference

Quickstart

Four commands take you from empty directory to a running agent on the runtime of your choice.

# 1. lay down the two required layers
strata init atlas-researcher

# 2. read the whole stack back — the core sample
strata core

# 3. validate: schema, bounds, and seam collisions
strata probe --strict

# 4. emit the native format for your runtime
strata surface --to claude > ./out/agent.md

strata.yaml

spec: 0.4
name: atlas-researcher
version: 2.4.0
model:
  preferred: claude-opus-4-8
  minimum: any-frontier     # degrade, don't fail
layers:
  seams: auto                # discover seams/*/SEAM.md
  sediment: append-only
imports:
  - git+https://github.com/vega/[email protected]
surfaces: [claude, openai, langchain, mcp]

Rules

Two files, or it doesn't resolve. strata.yaml and CORE.md are the contract; everything else is opt-in.
Bounds beat seams. If a seam asks for something BOUNDS.md forbids, probe fails the build rather than resolving it at runtime.
Imports are pinned. A bare branch reference is a warning; releases must pin a tag or a commit SHA.
Surfacing is lossy and says so. Every adapter reports what it dropped, so you never silently ship a weaker agent than you wrote.
Forks keep lineage. A forked agent records its upstream in assays/, so drift is measurable rather than anecdotal.

What probe returns

{
  "spec": "0.4",
  "resolved_layers": 8,
  "unpinned_imports": ["seam-scrape@main"],
  "bounds_conflicts": [],
  "surfaces_clean": ["claude", "mcp"],
  "surfaces_lossy": { "openai": "veins/ flattened to prose" }
}

On lock-in

The spec deliberately owns no runtime. If STRATA disappears tomorrow, the repository is still a readable, versioned description of an agent — which is the point. An agent you cannot read is an agent you cannot review, and an agent you cannot review is not one you should ship.

Lounge

Open floor for the people writing the agents — and the wall. Ship notes, spec arguments, and graffiti, all in public.

The wall

Open to everyone — no wallet, no sign-in. Press & drag anywhere in the dark to draw with dots, and the whole field sees it. Double-click wipes the wall (10s cooldown).

Runtime

Surfacing is the act of turning a STRATA repo into whatever your runtime actually eats. Every adapter reports what it had to drop, so a lossy target is a warning you read — never a silent downgrade.

Spec coverage · best adapter

96%

Surfaces run today

18,402

Median surface time

0.4s

Adapters

How much of the stack survives

claude

96%

mcp

88%

openai

74%

langchain

61%

system-prompt

38%

Known losses

What each adapter drops

Adapter	Layers kept	Dropped	Consequence
claude	10 / 10	—	Full fidelity, including veins and beds
mcp	9 / 10	beds/	Sub-agents flatten into the parent
openai	8 / 10	veins/, beds/	Routes become prose; ordering is advisory
langchain	7 / 10	veins/, beds/, assays/	No audit trail survives the export
system-prompt	4 / 10	everything executable	Identity and bounds only — the honest floor

Run strata surface --explain to see the loss report before you ship. Demo data.

Registry

Published agents, resolvable by name and version. Every entry is a real git repository — clone it, read the layers, fork it if it's close but not right. Click a column to sort.

Agent	Spec	Layers	Forks	Updated
vega.sol/atlas-researcher	0.4	8 / 10	312	2h ago
7Kdq…mPXa/ledger-clerk	0.4	6 / 10	271	9h ago
renji.sol/triage-nurse	0.3	7 / 10	240	2d ago
0xQuarry/schema-smith	0.4	5 / 10	190	4h ago
noor.sol/prose-editor	0.3	4 / 10	172	6d ago
atlas_mines/site-watch	0.2	3 / 10	41	3w ago

Agents on spec 0.2 still resolve — surfacing warns instead of failing. Demo data.

Seams

A seam is one skill: a folder with a SEAM.md, optional scripts, and a description precise enough that an agent knows when to reach for it. Seams are imported by reference and versioned independently of the agents that use them.

Anatomy

seams/diff-review/
  SEAM.md              # when to use it, and how
  review.py            # optional — seams may ship code
  fixtures/            # optional — examples to ground behaviour

The first line of SEAM.md is the trigger description. It is the only part always held in context — the body loads when the agent decides the seam is relevant, which is what keeps a 40-seam agent affordable.

Most imported

Load-bearing seams

#	Seam	Imports	Author	Version
1	seams/retrieval	2,140	vega.sol	1.4.2
2	seams/diff-review	1,786	0xQuarry	2.0.1
3	seams/cite	1,231	renji.sol	0.9.7
4	seams/redact	894	noor.sol	1.1.0
5	seams/sql	377	7Kdq…mPXa	3.2.0

Imports counted across published agents on spec ≥ 0.3. Demo data.