AI agent memory server
AnchorMind
An MCP server where agents store decisions, errors, and procedures as fragments and pull them back in the next session. Several agents share the same memory, with access split per token.
// Fortyslop / PAYCURR
We build systems that keep running.
System integration, legacy modernization, data processing, and persistent memory for AI agents. The person who designs it is the person who hands it over.
services/
| name | what we do | what you receive |
|---|---|---|
| si/ | Build a new system from requirements through deployment and operational handover. | Source, design docs, deploy and ops runbooks |
| legacy/ | Read an undocumented system and change it in steps, without stopping it. | Dependency map, migration plan, behavior comparison results |
| data/ | Build the path data travels: collect, clean, load, verify. | Pipeline code, reconciliation reports, runbook |
| memory/ | Add a memory layer so AI agents keep their context after the session ends. | Memory server setup, integration code, ops guide |
AnchorMind spec
The AI memory server we built and run ourselves, which the memory service is based on.
- Form
- MCP memory server + npm-installable stdio proxy
- Tools
- 20 (as of v5.12.0)
- Search
- Three layers: keyword → semantic → hybrid (RRF)
- Memory unit
- 7 fragment types: fact · decision · error · preference · procedure · relation · episode
- Access control
- Per-agent tokens, workspace isolation
How we work
- diag Diagnosis We listen, read the code and infrastructure, and write down the risks.
- proposal Proposal Scope, approach, and a schedule estimate with its reasoning, in writing.
- build Build We ship in small pieces and get your confirmation at each step.
- handover Handover Source, docs, and operating procedures, plus answers to your questions.
Systems we run ourselves
Deep research agent API
aiq
An API that takes a question, researches it in several steps, and writes a report. Long investigations run as separate jobs on a distributed executor.
Request a diagnosis
Your inquiry arrives by email. We read it ourselves and reply.