Persistent consulting memory
Persistent AI memory for consulting work
A chat forgets. A consulting firm cannot afford to. Solon maintains context around the firm, client, opportunity, user and artifact, so the same background does not have to be rebuilt in every prompt.
The context problem is not a prompt problem.
Better prompting does not fix it. The information a proposal depends on lives in a transcript someone else recorded, a methodology the firm approved two years ago, a decision made in a meeting and a constraint the client mentioned once. Assembling that by hand, every time, is the actual work.
What Solon maintains context on.
Six kinds of context, each with a different lifespan and a different owner.
Firm memory
- Approved methodologies
- Service descriptions
- Quality standards
- Reusable IP
- Brand guidance
Practice memory
- Domain expertise
- Specialist frameworks
- Team playbooks
- Preferred delivery approaches
Client memory
- Organisation context
- Relationships
- Prior engagement information
- Client-specific requirements
- Confidentiality settings
Opportunity memory
- Transcripts
- RFP
- Objectives
- Assumptions
- Open questions
- Decisions
- Proposed scope
User and role context
- Consultant role
- Partner responsibility
- Access rights
- Working preferences
- Review authority
Artifact context
- Current proposal
- Previous versions
- Approved sections
- Comments
- Intended audience
- Connected deliverables
Not every source should count equally.
Relevance is not the same as permission, and recency is not the same as authority. A recent client decision may take priority over an earlier assumption. A firm methodology may guide the approach without being copied verbatim. A previous client document may inform internal reasoning while remaining excluded from the client-facing artifact.
Prior engagement — knowledge policy
- Use for internal reasoning
- Allowed
- Reuse exact wording
- Not allowed
- Show source in client output
- Not allowed
- Reuse as general pattern
- Allowed
- Partner review required
- Yes
This is supported by a layered and weighted memory architecture that is being refined throughout the beta.
Where this is today.
The memory model is operational in the beta: context persists across sessions, sources are visible on the artifacts they informed, and reuse policies can be set per source. The weighting and retrieval behaviour is being refined with every engagement we run through it.
Honest status
- Context persists beyond a single sessionAvailable in beta
- Sources visible on the artifacts they informedAvailable in beta
- Per-source reuse policiesAvailable in beta
- Weighting between competing sourcesBeing refined
- Automatic retrieval across all firm systemsPlanned