Status: Available
L1 · Guardedelan1 Govern · L1 Guarded
“Can I let this run?”
The Right Path, approvals, a single writer and a tamper-evident record you can read.
L1 Guarded in detailGoverned Intelligence Platform
elan1 Govern: governance for every AI agent write
The governed write path for AI agents.
elan1 Govern is the control plane for AI agent writes. Every write intent climbs the Right Path: identity, nine pre-gates, versioned rules, a blast-radius guardrail and an autonomy check, then a named human approver where your rules ask. Each action, allowed or refused, leaves a 16-field, hash-chained proof record. A refusal costs 40% of one governed action.
Status: Available · L1 Guarded and L2 Attested
- KernelPort
- Six calls
- Right Path
- Rung 0b STOP to 9 METERED
- Proof record
- 16 fields, hash-chained
01What it does
One write function, reached through six calls.
A model may produce a write intent. It may never produce an allow.
elan1 Govern stands between any agent and your system of record. Every write intent climbs the Right Path. It passes nine pre-gates, versioned rules with no model, a blast-radius guardrail and an autonomy check. Where your policy asks, a named person approves. A single writer then makes the change. Govern writes a 16-field proof record and one meter event. Anything Govern refuses gets a typed code and a record.
Who it’s for
The people who answer for a write.
- Financial controllers and heads of internal audit who must show that an agent’s write could not have happened without the rule check and the approval.
- CFOs, who own the audit and compliance budget.
- The auditor’s counterparty, from L2: the external auditor, lender or insurer who receives the evidence.
- Agent vendors and ISVs who want their writes to run on one governed path.
02How it works
Every write intent climbs the Right Path.
Ten rungs, from STOP to METERED. Choose what happens at the gate and follow one write to its record, including the refusal.
ProposedMeter: Nothing counted yet
- 0bSTOP
- 1IDENTITY
- 2WRITE DOOR
- 3PRE-GATES
- 4evaluate()
- 5GUARDRAIL
- 6AUTONOMY
- 7ACT
- 8PROOF
- 9METERED
A write intent is proposed. Press Next rung, or the → key, to send it up the path.
Write intent: “Post purchase invoice · ₹4,82,000 · proposed by agent.” Illustrative specimen. Not customer data. You advance it; nothing plays on its own. Use the → key or the button.
Change the declaration. Watch the outcomes move.
Eight write intents from Kolam Components*, one declaration. The declaration decides, not a model: change it and see what is allowed, held for a named approver or refused.
Edit the declaration
Illustrative specimen — not customer dataDeclaration
decl_ap_purchase · v7 Kolam Components*
v7 · registered. Every record names the declaration and version it was checked against.
Illustrative specimen — not customer dataOutcome board · 8 write intents
Allowed1
Held for a named approver3
Refused4
1 governed action4 refusals at 0.4 governed action each3 held, not metered until decided.
The record this outcome writes · intent 2
Illustrative specimen — not customer dataL1 · Guarded
Proof record · Seal v6 · 16 fields “no seventeenth field, ever”
- Outcome
- Held for a named approver
- Deciding rung
- rung 6 AUTONOMY
- Approver
- usr_specimen_approver · named, not the asker
- Refusal code
- —
- Why
- Above the ₹1,00,000 ceiling.
- F01tenant_id
- tnt_specimen_kolam
- F02seq
- 2202
- F03prev_digest
- sha256:4be1…90c7
- F04write_id
- wr_specimen_p02
- F05write_type
- purchase_invoice.post
- F06declaration_id
- decl_ap_purchase
- F07declaration_version
- 7
- F08policy_version
- 2027.06.1
- F09actor_id
- agt_specimen_ap01
- F10actor_kind
- agent
- F11autonomy_level
- A4
- F12record_id
- pinv_specimen_0418
- F13record_version
- 0
- F14payload_hash
- sha256:c03a…11fe
- F15lexicon_hash
- sha256:e2c4…0b16
- F16recorded_at
- 2027-07-14T10:33:40+05:30
Held at rung 6 · obligation OB-specimen-0412 for usr_specimen_approver · not metered until decided
when · who · hash — checked and recorded by elan1 Governtamper-evident (L1) · No seventeenth field.
Illustrative specimen. Not customer data. *Kolam Components and Specimen supplier 07 are fictional; all data is illustrative. Codes are illustrative. The decision function runs in your browser; nothing is registered or sent.
In an accounting pack the same checks are named by what they check in the books (G1 Ledger … G4 GST); how the names map is an open question on the architecture page.
03How it connects
Govern governs the other three, and agents from outside.
Work and Decide reach Govern through KernelPort. Run carries Govern's work on five contracts (gateway, trace, tools, run and guard), and it asks Govern for a gate check at substitution step 7.3. Third-party agents connect over MCP or HTTP.
| Consumer | How it connects | Mode |
|---|---|---|
| elan1 Work | Through KernelPort. Every change and every undo climbs the Right Path. | Mode A where elan1 hosts ERPNext · Mode B on Attach |
| elan1 Decide | Through KernelPort. Only the writes that follow a decision reach Govern. | Mode B |
| elan1 Run | Carries Govern's work on five contracts: gateway, trace, tools, run and guard. Asks for a gate check at substitution step 7.3. | Mode B, failing closed |
| Third-party agents | Over the MCP server (34 tools) or the HTTP API. The in-process SDK gives Govern the write credential. | Mode B over MCP or HTTP · Mode A in process |
Wherever another surface shows a Govern decision, it carries one line: “Checked and recorded by elan1 Govern.”
04Key capabilities
Six modules on one write path.
Open a card to see the rung or the record field it maps to.
- 01
KernelPort
KernelPort has six calls: define, assert, seal, verify, scope, refuse. An MCP server (34 tools) and an HTTP API serve the agents you already run.
The six calls
- define
- Declare what may change, and through which write types.
- assert
- Propose a write intent. It starts its climb at rung 0b.
- seal
- Close the proof record for a write at rung 8.
- verify
- Check one record, or re-walk the chain.
- scope
- Bound what a caller may reach: tenant, write types, autonomy.
- refuse
- Return a typed code. A record is still written, metered at 0.4 of a governed action.
Available - 02
Right Path
Right Path runs from rung 0b STOP to rung 9 METERED. It includes pre-gates G1–G9, six of which can never be skipped. evaluate() applies versioned rules, with no model.
Rung mapping
- 0b STOP
- The stop switch is read first.
- 1 IDENTITY
- Who is asking, for which tenant, inside which scope.
- 2 WRITE DOOR
- The write type must be on the declaration's list.
- 3 PRE-GATES
- G1–G9 against your own data; six never skipped.
- 4 evaluate()
- Versioned rules. No model.
- 5 GUARDRAIL
- The blast radius must sit inside the declared limit.
- 6 AUTONOMY
- Above the ceiling, a named person decides.
- 7 ACT
- A single writer makes the change.
- 8 PROOF
- The 16-field record is written and hash-chained.
- 9 METERED
- One governed action counted; a refusal counts as 0.4 of one.
Available - 03
Declaration Studio
Declaration Studio holds your rules in seven versioned blocks. Every record names the declaration it was checked against.
Where it lands on the record
- Rung 2
- The write door opens only for declared write types.
- Rung 4
- evaluate() runs the declaration's versioned rules.
- F05
- write_type — the kind of write, from the declaration.
- F06
- declaration_id — the declaration that was checked.
- F07
- declaration_version — the version in force at the time.
- F08
- policy_version — the rules evaluate() ran.
Available - 04
Approvals
Approvals are named and step-up, and never come from the asker. Autonomy levels A0–A5 decide when a person must approve. Earned autonomy uses a deterministic score, with no model involved.
Rung and field mapping
- Rung 6
- AUTONOMY — above the ceiling, a named person decides.
- F09
- actor_id — the named approver, or the agent that proposed.
- F10
- actor_kind — human or agent.
- F11
- autonomy_level — A0 to A5, the level the write ran at.
Available - 05
Proof record and elan1-verify
The record has 16 fields under Seal v6, hash-chained per tenant. Evidence rows cannot be updated or deleted.
Field mapping
- Rung 8
- PROOF — the record is written for every outcome, refusals too.
- F01
- tenant_id — chains never cross tenants.
- F02
- seq — position in the chain; only ever goes up by one.
- F03
- prev_digest — the link to the record before.
- F14
- payload_hash — a hash, so personal data stays out of evidence.
- F16
- recorded_at — when the record was sealed.
Available - 06
Evidence (L2)
Evidence (L2) adds Ed25519 signatures, Merkle roots anchored to two RFC 3161 authorities within 15 minutes, an eight-member bundle and a free offline verifier.
What L2 adds, from L2
- Signature
- Ed25519 on each record, from L2.
- Anchor
- Merkle roots to two RFC 3161 authorities within 15 minutes.
- Bundle
- Eight members, handed to your auditor.
- Verifier
- Free and offline: checkable without us, from L2.
Available
05Three levels
Three levels of a claim. Each has its own page.
Levels describe what you can rely on. They are levels of a claim, not separate products or brands, and not tiers you buy.
- Status: AvailableL2 · Attested
elan1 Govern · L2 Attested
“Can I hand this over?”
A signature, anchoring, an evidence bundle and a free offline verifier: a signed receipt you can hand to your auditor.
L2 Attested in detail - Status: On auditor acceptanceL3 · Proven
elan1 Govern · L3 Proven
“Does this survive the vendor?”
A receipt that outlives the vendor.
L3 Proven in detailL3 begins when an outside auditor accepts an L2 bundle in writing.
06Inside the product
The Govern console, screen by screen.
What an approver, a controller and an auditor see: the day's queue, a held write, a refusal explained, the proof record and the chain. Choose a screen; each lists what to look for.
01 / 14 · Work
Today
- Four counts that matter: waiting on you, expiring within the hour, not done, and statutory dates due.
- Each held write shows who proposed it, how many checks passed and the time left to decide.
- Refusals are counted here too, each metered at 0.4 of a governed action.
02 / 14 · Work
Approvals inbox
- Agents and people propose; only people approve.
- There is no bulk approve: one decision per obligation, against the diff you were shown.
- Expiry comes from the value declared for each write type, not a global timer.
03 / 14 · Work
Deciding a held write
- Before and after, field by field; material changes are marked and nothing is summarised away.
- Declared blast radius, reversal plan and deadline sit above the decision.
- The approval is recorded as its own governed write, linked to the record, the proof and the rules applied.
04 / 14 · Work
Right Path rung trace
- This stock move stopped at rung 3, pre-gates; rungs it never reached say Not reached, never blank.
- Rung 8 still recorded a proof, and rung 9 metered it at 0.4 of a governed action.
- Declaration, policy version, channel and reversal are on the same page as the outcome.
05 / 14 · Work
A refusal, explained
- Both results side by side: 4,800 kg in the sandbox, 3,650 kg on live two seconds later.
- The snapshot was restored and the next step is stated in plain words.
- The refusal has its own proof record and is metered at 0.4 of a governed action.
06 / 14 · Assurance
Refusals
- 171 refusals split into gate, policy and our own platform failures.
- Every gate and policy refusal names the rung it stopped at and what it cost: 0.4 of a governed action.
- Our own platform failures are shown separately, paged, and never billed.
07 / 14 · Assurance
Proof record
- Each record carries the digest of the one before it, so the chain can be re-walked.
- The screen says what it cannot claim: hash-chained, not signed, at Assurance Level 1.
- Enforcement mode, channel and autonomy level are recorded on the entry itself.
08 / 14 · Assurance
Chain standing
- A nightly full walk plus a re-walk of the head every few minutes, each printed with its scope.
- Refusals and waiting writes sit in the same chain as committed ones.
- The scope is stated, not implied: 19 of 23 write types.
09 / 14 · Assurance
Enforcement by channel
- Enforced means the kernel holds the only credential to the system of record.
- Attested advisory means the caller holds the credential, and the screen lists what you may not say.
- A dash means the route cannot attempt that action at all.
10 / 14 · Records
Declarations
- Every declaration has a signed-off version and a named person who signed it off.
- Drafts are visible and stay out of production until signed off.
- Changes are compared version to version, never edited in place.
11 / 14 · Agents
Autonomy ladder
- A missing grant reads A0 Observe, never a blank.
- Raising a grant is itself a governed write with dual control; taking it back is one press.
- No reversal plan means no A3 or above.
12 / 14 · Agents
Agent map
- Write-capable edges are drawn heavy; staging and read-only edges are dashed or dotted.
- Only act.write can change a record, and it lands in the system of record.
- People, not agents, sit in the resolver sets.
13 / 14 · Work
Stop switch
- Reads, verification and exports keep working; a stop holds writes, not reads.
- Open approvals freeze with their expiry paused.
- The stop is verified by attempting a write and expecting the refusal.
14 / 14 · Assurance
Meter
- Refused writes appear on the same table at 0.4 of a governed action each; escalated writes appear at quantity zero.
- The meter is reconciled against the proof chain nightly: zero mismatches.
- The billing basis comes from the declared blast radius and is the same for every customer.
Illustrative specimen. Not customer data. Screens come from the elan1 Govern console running on fictional sandbox data.
07Assurance level and mode
Say only what the mode allows.
L1 · Guarded
Available now.
L1 Guarded, with L2 Attested available. Mode A, enforced, where Govern holds the write credential through the in-process SDK. Mode B, attested advisory, over MCP or HTTP: evaluated and recorded.
Mode A · Enforced
Govern holds the credential
Through the in-process SDK, Govern holds the only credential to the system of record. No write reaches it without a gate decision and a proof record.
- We sayEnforced. Tamper-evident to you at L1.
- We never saySigned or checkable offline at L1. Those come with L2.
Mode B · Attested advisory
Your system holds the credential
Over MCP or HTTP, your system makes the write. Govern evaluates it against your rules and records the result.
- We sayEvaluated and recorded.
- We never sayEnforced, blocked or stopped.
L2 Attested is available: signed records, checkable offline. L3 Proven is reached when an outside auditor accepts an L2 bundle in writing. How the levels work.
08Meter and pricing
Paid per governed action. A refusal costs 40% of one.
The unit
One governed action
One committed write. A refusal is metered at 0.4 of one.
The unit is one committed write. A refusal costs 40% of one: it still runs every check and writes a record, but nothing is written to your books. There is no per-seat or outcome pricing. From L2, evidence is billed as its own line.
- A yearly platform floor per tenant, set by assurance level, with a governed-action allowance included and consumed first.Excl. GST
- Actions beyond the allowance are billed per action, at the rate for your level. All prices are exclusive of GST.Excl. GST
- A 90-day ledger1 pack pilot through your CA or audit firm.Excl. GST
Platform floor by assurance level
Levels describe what you can rely on. They are levels of a claim, not separate products. Each floor includes a yearly allowance of governed actions, consumed first.
| What | Price · excl. GST | Includes |
|---|---|---|
| L1 Guarded | ₹1,80,000 per tenant per year | 4,80,000 governed actions a year included |
| L2 Attested | ₹6,00,000 per tenant per year | 15,00,000 governed actions a year included |
| L3 Proven | ₹24,00,000 per tenant per year | 42,00,000 governed actions a year includedApplies once an outside auditor accepts your L2 bundle in writing. |
| Sovereign or on-premises deployment | ₹1,20,00,000 per deployment per year | Your own residency domain and keys; the floor carries the fixed cost |
Governed actions beyond the allowance
Only actions beyond the included allowance are billed. A refusal counts as 0.4 of a governed action: it uses 0.4 of an included action while the allowance lasts, and is then billed at 40% of your level's rate. It still runs every check and writes a proof record.
| What | Price · excl. GST | Includes |
|---|---|---|
| Beyond allowance, L1 Guarded | ₹1.60 per governed action | |
| Beyond allowance, L2 Attested | ₹1.20 per governed action | |
| Beyond allowance, L3 Proven | ₹0.90 per governed action | |
| Sovereign deployment | ₹0.55 per governed action | |
| Refusal, at any level | 40% of one governed action | Within the allowance, a refusal uses 0.4 of an included governed action; beyond it, it is billed at 40% of your level's rate |
| Refusal beyond allowance, L1 Guarded | ₹0.64 per refusal | 40% of ₹1.60 |
| Refusal beyond allowance, L2 Attested | ₹0.48 per refusal | 40% of ₹1.20 |
| Refusal beyond allowance, L3 Proven | ₹0.36 per refusal | 40% of ₹0.90 |
| Refusal, sovereign deployment | ₹0.22 per refusal | 40% of ₹0.55 |
Add-ons
| What | Price · excl. GST | Includes |
|---|---|---|
| Regulatory pack (ledger1 · invoice1 · collect1) | ₹1,35,000 per tenant per year | Configuration of Govern for one rule set; blended price |
| Evidence bundle beyond allowance | ₹2,500 per bundle | From L2: the eight-member signed bundle |
| Verifier profile | ₹3,00,000 per regime per year | From L2: offline verification profile for one regulatory regime |
| Counterparty verification seat | ₹45,000 per seat per year | For an auditor, lender or insurer who checks your bundles |
| Partner certification (CA or audit firm) | ₹1,50,000 per firm per year | Introducer, Implementer or Verifier partner |
All prices are exclusive of GST. Prices are current list prices; GST is added at the applicable rate on the invoice.
Four meters, one per layerExclusive of GST
- elan1 Workcredit₹0.80 per creditbeyond the plan allowance
- elan1 Decidedecision₹30 per 1,000 decisionsbeyond the plan allowance
- elan1 Governgoverned action₹1.60 · ₹1.20 · ₹0.90 per governed actionL1 · L2 · L3, beyond the allowance
- elan1 Runnode-year₹1,80,000 per node per yearand the same per GPU, plus the subscription
Refusals: a refusal by elan1 Govern costs 40% of one governed action (₹0.64 at L1 beyond the allowance) and still writes a proof record. A call refused at the edge by elan1 Run costs the same 40% (₹0.64 at the L1 rate), and no token is sent.
elan1 Govern floors, per tenant per year: L1 ₹1,80,000 · L2 ₹6,00,000 · L3 ₹24,00,000 (L3 once an outside auditor accepts your L2 bundle). Exclusive of GST.
09Where it is weak
What elan1 Govern does not do.
Every claim on this page has an edge. These are ours.
- Mode B records the write; it cannot stop it.Over MCP or HTTP your system holds the credential. Only Mode A, through the in-process SDK, is enforced.
- At L1, your auditor still needs us.Records are tamper-evident to you, not signed. For offline checking without us, choose L2 Attested.
- An outage stops allows.If elan1 is unreachable, changes wait for a person. That is the safe direction, and it is slower.
- It does not make you compliant.It records who allowed each write and under which rule. Classification, conformity work and model testing stay with you.
10Status
Available now: L1 Guarded and L2 Attested.
Both levels are available. L3 Proven is reached on an event, never on a date.
- KernelPort v1Six calls, frozen, with the MCP server (34 tools) and the HTTP APIAvailable
- L1 GuardedThe full Right Path, approvals, a single writer and a tamper-evident proof recordAvailable
- L2 AttestedEd25519 signatures, anchoring to two RFC 3161 authorities, the evidence bundle and the free offline verifierAvailable
- L3 ProvenWhen an outside auditor accepts an L2 bundle in writingOn auditor acceptance
Available now. Future improvements are listed, in order and without dates, on What’s next.
Announced 10 Oct 2026: elan1 Govern, Work, Decide and Run are available.
11FAQ
Nine direct answers.
Most AI governance platforms inventory models and manage policies on paper. elan1 Govern works at the write itself: each change an AI agent proposes is checked against versioned rules, held for a named approver where your rules ask, and recorded in a 16-field proof record. It governs actions, not only documents.
No. A model may produce a write intent, but it may never produce an allow. No model sits on the allow path, and an approver is never the person or agent that asked.
The agent gets a typed refusal that names the rung and the reason, so its code can handle it. A proof record is still written, and the refusal is metered at 0.4 of a governed action, less than an allowed write.
That depends on the mode, and we always say which applies. In Mode A, where Govern holds the write credential through the in-process SDK, your rules are enforced. Over MCP or HTTP (Mode B), your system holds the credential, so Govern evaluates and records each write but cannot block it.
It helps produce evidence: a record of every agent action, the rule it was checked against and the named person who approved it, which supports record-keeping and human oversight duties. No product makes a system compliant on its own; classification, conformity work and model testing stay with you.
A yearly platform floor per tenant, set by assurance level: ₹1,80,000 for L1 Guarded and ₹6,00,000 for L2 Attested, each with a governed-action allowance. Actions beyond the allowance cost ₹1.60 each at L1 and ₹1.20 at L2. A refusal costs 40% of one governed action: ₹0.64 at L1 and ₹0.48 at L2 beyond the allowance. All prices are exclusive of GST.
Yes. L1 Guarded and L2 Attested are available now. Early adopters can also join the design-partner programme and shape what comes next.
Allows stop, and changes wait for a person. Stopping allows is the safe direction.
At L1, records are tamper-evident to you. Offline checking without us comes with L2 Attested, which is available now.
Next steps
Related reading
AI agent governance: a practical guideInventory, write paths, rules, approval and evidence, with a 30-day plan.Human-in-the-loop for AI agentsWhen to ask a person, who that person is and what they must see.AI agent audit trail, field by fieldWhat to record for each action, refusals included.EU AI Act Articles 12 and 14Logging and human oversight when AI agents act.AI agent audit trail: the proof recordSixteen fields, hash-chained, tamper-evident at L1.Assurance levelsL1 Guarded, L2 Attested, L3 Proven.Reference architecturePlanes P0–P8 and the two seams.What we claimEvery claim with its level and status.
One write path, four layers
Proposeselan1 WorkProposes what should changeProposeselan1 DecideProposes the answerControl planeelan1 GovernAllows, refuses, recordsRuntimeelan1 RunCarries every call under contract
Work and Decide propose. Govern and Run assure. Only Govern allows.
// elan1 GOVERN
See one write climb the Right Path.
Follow a governed action from write intent to proof record, and watch the refusal branch write its own record, metered at 0.4 of a governed action.