Per-matter ACL enforcement
Aeira's 8-axis ACL pre-filter enforces matter membership at the index, not at the application layer. A user not assigned to the matter cannot retrieve from it. The rows do not enter the candidate set.
Sparcle for Legal
Privileged content stays inside the matter.
Privilege is non-negotiable. Conflicts walls are non-negotiable. Bolt and Aeira ship for that: per-matter ACLs enforced at the index, BYO LLM with on-prem inference, audit chain your partner-in-charge can read end-to-end. The verifier binary is the artifact for client AI-disclosure requirements.
Compliance frameworks
What Sparcle ships against.
Legal compliance is the intersection of attorney-client privilege, conflicts of interest, ABA Model Rule 1.6 and state analogs, and the emerging state bar AI-disclosure rules. Each row below maps to a concrete implementation.
| Framework | What Sparcle ships against it |
|---|---|
| Attorney-client privilege + work product | Privilege protection is non-negotiable. Bolt's per-matter ACL enforcement, identity-bound retrieval, and audit-trail responses ensure privileged material doesn't leak across matters, and Aeira seals every search it serves into the audit chain for partner-in-charge review. |
| Conflicts of interest | The Authority Policy SDK lets the conflicts team encode firm-specific conflict walls at the retrieval layer. Conflicts are enforced before retrieval, not after the fact. Conflict-wall enforcement is exercised by the test suite, not left to hope. |
| ABA Model Rule 1.6 + state analogs | Confidentiality of client information across electronic systems. Bolt masks structured identifiers before any LLM call. For privileged content, run your own model (Mode A or B) so it never reaches a vendor cloud; the BYO-LLM mandate gives the firm direct control over the inference endpoint. |
| State bar AI-disclosure rules (2024+) | California, Florida, New Jersey, and others now require disclosure when AI is used in legal work. The audit chain is the artifact the firm produces: every AI invocation has a record, and the standalone verifier confirms integrity. |
Data residency posture
Privilege boundary, enforced mechanically.
BYO LLM + on-prem inference path
Bolt runs against your LLM endpoint: on-prem Llama, Mistral, Qwen, or Azure OpenAI inside the firm's tenant. Privileged content does not leave the firm's network in the supported postures.
Customer-held KMS
Vault Transit, AWS KMS, Azure Key Vault, or PKCS#11 HSM. The firm holds tenant master keys; Sparcle does not. Crypto-shred is per organization; matter-level erasure is a scoped delete.
Audit chain the partner-in-charge can read end-to-end
Every Aeira search, every tool invocation, every human approval, every masking and
authorization decision, and the security-relevant identity transitions that open and
close access to a matter, sealed and independently verifiable. The audit chain is the
partner-in-charge's view of what happened on a matter, without requiring the firm to
take Sparcle's word for it.
1-page brief
Brief in preparation.
Workflows the founding team has built for
Four anchor use cases.
The workflows below are scoped with legal design partners. The legal Authority Policy pack and matter-ACL enforcement are the per-firm tuning points.
- Diligence review with conflict walls Diligence-corpus retrieval respects matter membership and conflict walls. The Authority Policy ranks current, signed-off materials above unsigned drafts.
- Drafting assist with privilege preservation Draft assembly pulls from the matter's own corpus only. Privileged material does not cross matter boundaries; every search behind the draft is sealed into the audit chain for partner review.
- Knowledge management across matters Firm-wide KM lookups respect per-matter ACLs while letting partners search across closed matters they have access to. The Authority Policy pack for legal handles practice-area evidence ladders.
- Client disclosure of AI use Time-bounded audit-chain export per matter, with the standalone verifier binary. The firm produces the verifier output for client-disclosure requirements; the client can run the verifier themselves.
Trust dossier for legal
The four artifacts your CIO and partner-in-charge will ask for.
/trust/where-the-model-runs
How structured identifiers are masked before any outbound LLM call; how Mode A (on-prem) and Mode B (in-tenant cloud) preserve privilege.
/trust/verify
The standalone audit-chain verifier. Your IT runs it; the partner-in-charge sees the output. The artifact for client AI-disclosure requirements.
/trust/subprocessors
Asymmetric sub-processor topology. Why client matters do not transit Sparcle infrastructure in the Helm or air-gap topologies.
/trust/incident-response
Five-phase response framework. The runbook a partner-in-charge expects when a security event involves client-confidential material.
Want to walk this against your specific privilege and conflicts posture?
A 30-minute legal architecture review covers your matter-ACL needs, your conflicts walls, your LLM endpoint posture, and the workflows your practice groups need. Founder-led; the engineering team is on the call.