Coverage That Makes Agents Decisive
Conyr turns first-party Solana activity into decision-ready actor intelligence: position-correct wallet skill, funding provenance, behavioral similarity, coordinated-wallet evidence, bundle lifecycle, and exit-aware liquidity. An agent gets normalized conclusions with evidence and explicit confidence states—not a transaction dump it has to interpret from scratch.
25 registered trading programs · 20-hop provenance · 128 curated origins · 30-dimensional behavior · 10 bundle structures
Conyr’s coverage contract is an agent-safety feature: supported evidence becomes structured intelligence; unavailable or incomplete evidence stays explicit instead of becoming a plausible-looking guess.
First-party ingestion and normalization
Conyr receives Solana transactions through self-hosted Yellowstone gRPC infrastructure and converts recognized trading activity into evidence-linked product records. The active parser registry spans 25 trading program IDs, with venue-specific instruction decoding for core venues and normalized flow analysis across the wider supported surface.
For an agent, that means:
- one normalized trade shape across supported venues;
- transaction identity, slot, and event time retained for audit;
- immediate downstream updates to positions, PnL, labels, provenance, bundles, and coordination state;
- unsupported or malformed evidence represented as unavailable instead of assigned invented semantics.
The production pipeline has demonstrated more than 20 million decoded swaps per day in a dated June 2026 observation window. Use the claim registry for the measurement rule and citation.
Raw activity is the input, not the product
FIFO inventory, realized outcomes, remaining position state, hold time, and entry/exit evidence—so a profitable partial exit is not mistaken for a won position.
Trace inbound capital through as many as 20 hops toward curated exchange, custody, payment, and infrastructure origins.
Compare eligible wallets across 30 explainable dimensions spanning timing, exits, performance, sizing, infrastructure, and provenance.
Retain directed follow relationships, campaigns, entities, crowd state, bundle membership, and evidence instead of flattening them into one opaque score.
Combine bundle structure, holder concentration, wash patterns, funding abuse, LP state, and coordinated-dump impact for one agent decision.
Query the same intelligence through 46 MCP tools, 41 typed REST operations, and 13 WebSocket channel families.
Freshness by intelligence layer
A streaming input does not give every derived aggregate the same clock. Agents should use these operating semantics and reconcile stream events with REST snapshots.
| Intelligence | Normal update behavior | Agent behavior |
|---|---|---|
| Decoded swaps and trades | As transactions are processed | Use signatures, slot, and event time as the evidence boundary |
| Positions and realized PnL | As qualifying trades update inventory | Preserve partial or unavailable cost basis |
| Funding provenance | As qualifying funding transfers are processed | Preserve unresolved paths rather than inferring an origin |
| Public identity registry | Direct /identity and /kol-wallets calls read active Postgres rows at request time | Use the direct routes when current aliases matter; preserve registry provenance and timestamps |
| Wallet labels | Recomputed as qualifying evidence changes; compact KOL fields use a 30–60 second dictionary projection and the response can be cached for 60 seconds | Retain label timestamp and confidence; do not use label absence to override direct identity |
| Behavior vectors | Target refresh approximately every 60 seconds | Treat no vector as unavailable, not a low score |
| Follow graph and campaigns | Near-real-time scored updates | Read support, direction, and degraded state |
| Bundle lifecycle | Hot-path detection and status updates | Reconcile lifecycle and inventory from current state |
| Published scale measurements | Updated when formally measured | Treat them as dated evidence, not dashboard counters |
Decision-safe response semantics
Conyr preserves absence states because they lead to different actions:
| Representation | Meaning for an agent |
|---|---|
null / unavailable | The current evidence cannot support a value. Do not coerce it to zero or “clean.” |
| Empty list | The request succeeded and no retained records met the documented scope and threshold. |
| Zero | The quantity is in scope, measured, and equal to zero. |
partial: true | Useful sections returned while one or more named dependencies were unavailable or inapplicable. |
| Unknown provenance | No recognized origin was resolved on the available path. It does not mean self-funded. |
| No behavior vector | The wallet did not pass the activity gate or a current vector was unavailable. It is not a low score. |
degraded: true | Coordination results are usable with reduced source completeness; inspect the accompanying state. |
primary: null, aliases: [] | The syntactically accepted wallet identifier has no active public-identity registry match. It is not proof of anonymity or safety. |
wallets: [] | The valid normalized handle and optional platform filter matched no active wallet records. Do not silently broaden the platform. |
Public identity is attribution, not ownership proof
Direct KOL identity responses carry verification: "source_asserted", the registry source, and record timestamps. This means the source asserted the wallet-to-handle association. It does not mean the wallet signed a challenge, the current social-account owner still controls it, Conyr endorses the actor, or the wallet is skilled or safe.
Use identity for display and discovery. For a consequential conclusion, keep the address visible, re-read current aliases, and corroborate with transaction signatures, position-correct quality, funding provenance, entity membership, follow evidence, and bundle or campaign evidence. See KOL Identity Workflows for the endpoint-selection and join contract.
Wallet and token coverage
Conyr’s latest published production measurement observed more than 3.4 million active wallet addresses and 660,000 traded token mints. The corpus continues to change with the chain. Wallet and token endpoints layer market activity with position state, labels, provenance, bundle membership, holder behavior, liquidity, and security evidence.
Address-level intelligence stays evidence-first: an address may be automated, custodial, or one of several wallets controlled by an actor. Use Conyr’s entity, bundle, follow, and provenance surfaces when the decision requires actor-level context.
Position-correct PnL
Conyr reconstructs FIFO cost basis and position state before aggregating wallet outcomes. Partial exits consume inventory instead of becoming independent wins, and entry/exit signatures remain attached to the result. The /quality surface reports closed-lot win rate beside raw exit-row win rate and names the inflation gap so an agent can distinguish repeatable position outcomes from exit style.
When transfer-acquired inventory or incomplete history prevents a supported cost basis, Conyr preserves that state rather than fabricating profit.
Behavior fingerprints
Eligible wallets receive a 30-dimensional fingerprint across timing, exit discipline, performance shape, velocity and sizing, infrastructure, and provenance. The current activity gate requires at least 10 closed positions, one day of activity, and 15 trades. This deliberate gate keeps thin histories out of similarity search instead of turning weak evidence into a confident profile.
Similarity describes behavioral resemblance. For control or coordination questions, combine it with entity, bundle, funding, and repeat follow evidence.
Funding provenance
The resolver walks capital backward through as many as 20 hops toward 128 curated exchange, custody, payment, service, and other recognized infrastructure origins. Results retain the root, path, and hop count when supported. Shared infrastructure is treated as provenance context—not automatic proof that every funded wallet belongs to one operator.
Bundle intelligence
The bundle engine represents 10 structural classifications across execution order, launch cohort, funding, synthetic behavior, and hybrid patterns. It keeps structure, lifecycle, member inventory, supply at risk, realized PnL, and weighted reason-code evidence available to the agent. Some product surfaces consolidate those structures into five headline risk archetypes for faster routing.
Coordination and actor resolution
Follow edges retain direction, repeat support, lag distribution, and trade-size resemblance. Campaigns represent time-bounded token episodes; entities aggregate stronger shared-funder or shared-execution evidence; crowd state shows how those actors are moving now. Known public infrastructure is suppressed where it would otherwise create a false group.
Conyr reports observed structure and its evidence. The agent decides how that evidence affects its policy; it should not invent off-chain identity, intent, or certainty the chain cannot prove.
Operating boundaries agents must preserve
- New venue deployments and changed instruction shapes can precede normalized support. Check the returned evidence state, not a remembered venue list.
- A position or funding path that begins before available history may remain partial even when recent activity is complete.
- Behavioral classifications evolve as evidence accumulates; retain timestamps and confidence when caching them.
- Published scale values carry explicit as-of dates and counting rules in the claim registry.
- Markets are adversarial. Combine signals across evidence domains and reconcile live events with current state before acting.
These are execution rules for reliable autonomous decisions—not reasons to reduce Conyr to a raw data feed. The differentiation is the intelligence Conyr reconstructs above the chain and the evidence it preserves beneath every answer.