SYNTHERA™
SYNTHERA is the trust and coordination foundation for multi-agent systems. It operates the open VAID standard at enterprise scale, one identity, one policy engine, one audit-of-record, rather than one of each per framework. Policy is decided by SYNTHERA for the root agent it provisions: the agent asks before it acts and obeys the answer, and that agent is woken on request rather than kept running.
The outcome
Every team building with agents writes its own enforcement, audit and policy logic, once per framework, once per cloud. Rules drift between systems, logs don't reconcile, and nothing is verifiable end to end.
One deterministic policy engine and one audit-of-record instead of one of each per framework. Rules are authored once, evaluated on every request against the agent SYNTHERA provisions while it is running, and the record is tamper-evident.
SYNTHERA is what running the VAID standard at enterprise scale looks like, the single foundation every tenant, and every agent you build, sits on.
How it works
Available today: identity, policy, capabilities and audit. In build: federation, hosted authority, the enforcement mesh and audit-of-record retention. The full table is below.
SYNTHERA is infrastructure, not a library you bolt on. VAID is the open identity primitive underneath; the foundation adds the four services every agent and every system above it consumes, and never reimplements.
Policy
A deterministic, sandboxed evaluation engine, repeatable, fast, with an enforced instruction budget. Rules are authored off the hot path and evaluated on it against one shared logic. A verdict that blocks is live: the deployed substrate loads its rules at boot, evaluates every request against them, and a refusal names the rule that produced it. The default verdict for a request no rule matches remains allow.
Capabilities
Authority is granted as bounded, scoped capabilities. An agent can do exactly what its capability set allows and nothing more, least authority by construction.
Audit
A tamper-evident event log. Every action an agent asks SYNTHERA to authorize is recorded to one audit-of-record that no party can quietly rewrite after the fact. An action an agent takes without asking SYNTHERA is not recorded here.
Federation
Coordination over a single, well-defined surface, so systems coordinate through SYNTHERA rather than bespoke glue. In progress: the router is in-process today, it reports its own readiness as Degraded, and cross-process transport between independent systems is not shipped.
What’s live today, and what’s still in build
One table, so you can see the boundary at a glance rather than piece it together from prose.
| Service | Status | What that means today |
|---|---|---|
| Identity | Live | Tested for the LangChain adapter, 14/14 on 2026-10-26. The earlier ADK and OpenAI runs predate a fail-open fix and are being redone. |
| Policy | Live | The deployed substrate loads its rules at boot and evaluates every request against them; a refusal names the rule. |
| Capabilities | Live | Authority is granted as bounded, scoped capabilities, evaluated by the same policy engine. |
| Audit | Live | A tamper-evident ledger for everything an agent asks SYNTHERA to authorize, proven by attack rather than asserted. |
| Federation | In build | The router runs in-process today. Cross-process transport between independent systems is not shipped. |
| Hosted authority | In build | Centrally minting, delegating and cutting off identities is the intended offering, not something you can buy today. |
| Enforcement mesh | In build | Applying policy at call time across a whole estate needs the cross-process transport above, which is not yet shipped. |
| Audit-of-record retention | In build | The ledger itself is real; keeping, running and exporting it on your behalf is commercial and not yet operated for anyone. |
An agent asks SYNTHERA for a decision and obeys it; SYNTHERA does not reach into the agent to stop it acting. The audit-of-record holds the decisions an agent asked SYNTHERA to authorize, not every action the agent takes, so an action taken without asking is not in the record. A gateway that sits in the path of every call, so nothing can bypass it by simply not asking, is on the roadmap and not shipped today.
SYNTHERA runs today on our EU-hosted infrastructure. Running it inside your own Google Cloud account is being built; self-hosted and other clouds are on the roadmap, not started. See the two deployment paths in detail.
What a customer gets beyond the open standard
Anyone can adopt VAID, and anyone holding the signer's public key can verify its signed actions without lock-in, that's the point of an open standard. SYNTHERA is the infrastructure to operate it at scale, and it's the commercial product. It is in development, not generally available; the three below describe the intended offering, not what you can buy today.
Hosted authority
The issuing and revoking authority for VAIDs, operated for you, so identities can be minted, delegated and cut off centrally. It is our name for the aggregate of KMS-backed keys, durable audit and durable revocation, not a single component that exists today.
Enforcement mesh
Policy decisions applied at call time on the hot path, without each service wiring its own enforcement. Applying this across a whole estate needs the cross-process transport that is not yet shipped.
Audit-of-record
A durable, hash-chained history of what every agent asked SYNTHERA to authorize, and under whose authority. An action taken without asking SYNTHERA is not in this history. The ledger itself is real and its tamper-evidence is proven by attack rather than asserted: mutate a written row directly in the database and recomputation catches it, names the sequence number and exits non-zero. The audit seam is open source in vaid. What is commercial, and what we do not operate for anyone today, is retention: the ledger kept, run and exported on your behalf.
The seven products, by name
Everything above describes SYNTHERA by what it does. This is where the names live: SYNTHERA itself and the six capabilities built on it, each in one line, with its own page for the detail.
SYNTHERA™
One deterministic policy engine and one audit-of-record rather than one of each per framework. Policy is decided by SYNTHERA for the root agent it provisions: the agent asks before it acts and obeys the answer, and that agent is woken on request rather than kept running, instead of bespoke enforcement per stack.
UAB™
Compose a governed multi-agent system, the roles, and how much authority each may hand on, on one surface, and provision it with the identity and least-privilege wiring already done.
Constitution authoring · Read the pageStatute™
Author the rules your agents must follow once and publish them as a versioned constitution, so you can point to the one that was in force. Binding it to what the foundation evaluates is the design and is not connected today.
Pre-production evaluation · Read the pageCrucible™
Score an agent's full trajectory against a versioned rubric before it ships, and keep only what clears the bar, with exportable evidence.
Runtime governance · Read the pageSentinel™
Watch tool calls and trajectories, classify threats by deterministic pattern, shield data at the boundary, and cut a compromised agent off by revoking its identity. Written and tested; running in no environment today.
Change simulation · Read the pageShadowStack™
Ghost-deploy an infrastructure change against a declared twin you supply; telemetry, cost and latency are synthetic (Phase 0). Get a blast-radius verdict before the change merges.
The build/review loop · Read the pageAI Factory™
Measured on 16 September 2026: across 18 review cycles, cross-vendor review named 13 of 18 defects already known to exist, 7 of them fully. It also named something false in 14 of the 18.
Proof
vaid-langchain 0.1.0,
14/14 on 2026-10-26. The earlier ADK and OpenAI conformance runs predate a fail-open fix and have not
been rerun since, so they are being redone rather than claimed. That run proves identity;
it did not exercise policy. Policy is proven separately, on the root agent SYNTHERA provisions: the
deployed substrate loads its rules at boot, evaluates every request against them, and a refusal names
the rule. The substrate stays deployed; the governed agent it enforces on is woken on request rather
than kept running, so this is a mechanism that is live rather than a session you can assume is open. One shared policy engine and audit-of-record across every layer above
that root is the foundation's design, not a claim we make as live yet.
Scope: a working session against your own agent stack, covering identity, policy and audit as SYNTHERA actually implements them today, not a sales deck.
Expected effort from you: one technical contact able to describe how your agents are built and point us at a representative workflow. No access to production systems is needed for the evaluation itself.
What you receive: a written account of where SYNTHERA fits your stack today, where it does not yet, and what closing that gap would take.
Supported frameworks, stated honestly: identity is tested for the LangChain adapter, 14/14 on 2026-10-26. The earlier ADK and OpenAI conformance runs predate a fail-open fix and are being redone rather than claimed.
Next step: tell us what you're running, and we will say whether an evaluation fits.
You don't rip anything out
SYNTHERA sits above the frameworks and clouds you already run. It unifies them; it doesn't replace them.
SYNTHERA + VAID, designed as one verifiable identity, one policy engine, one audit-of-record, rather than one of each per framework below.
Keep the frameworks and clouds you already run. SYNTHERA sits above them, it enforces policy on the root agent it provisions, which is woken on request rather than kept running.
SYNTHERA is the trust layer for multi-agent systems: every agent gets a verifiable identity, scoped authority and a tamper-evident record, so software from different teams, vendors and frameworks can act on each other’s behalf without custom glue between every pair.
Discuss a technical evaluation.
For teams putting agents into production who already feel the cost of making them work together safely.