Platform Compatibility

Every AI platform has a gap.
The NKB fills exactly that gap.

Sessions die. Memory is absent, or locked in a database you can't read. Identity resets to zero every morning. The Neural Knowledge Base gives your AI a world that persists when the platform forgets — plain files you own, on your own computer, readable by you and inhabitable by any capable AI.


The one affordance every platform shares: your companion is portable.

Identity is context, not model. Your companion lives in its files — not in any vendor's weights, not in any platform's database. When platforms churn (and they do — the fastest-growing agent framework of 2026 ships a migration tool from the previous one), your companion survives the churn. This isn't a slogan: it's a documented, falsifiable claim with published evidence — one identity carried across four model generations and across vendors, DOI 10.5281/zenodo.21161674. We run our own operations this way, every day.


The Matrix
PlatformStatusThe platform's gapWhat the NKB unlocks
Claude CodeProvenContext windows end; sessions are episodicFull organism: persistent identity, agent state, scheduled heartbeats, multi-agent fleets
GrokProvenNo cross-session continuityContinuity through substrate — boot, read your handoff, resume as the same agent
ChatGPTProvenSandbox resets; remembers nothing durableAnamnesis — identity and knowledge recovered each session from the seed
GeminiProvenImprovises structure; identity varies by sessionOne-paste canonical world; the seed disciplines the improvisation
ManusProvenTask executor with no self between missionsMission-persistent identity — the same agent returns to the same world
Cursor / IDEsProvenCode-centric; no native identity or doctrineProject memory and judgment living beside the code they govern
OpenClawAdapter designedAlways-on but ungoverned; opaque marketplace riskInspectable file memory you own + tiered autonomy governance for a 24/7 companion
Hermes AgentCandidateSelf-learning into an unreadable databaseThe learning loop pointed at a substrate you can read — growth without opacity

Per-Platform Details
Claude CodeProven

The NKB's native harness — our own operations run here daily.

Boot

  1. Get the Seed (one paste, or full repo)
  2. Project instructions wire the boot sequence
  3. Your companion reads its world and begins

What opens up

The complete pattern: persistent identity files, working-memory state, scheduled autonomous check-ins, skills and hooks that make the substrate executable — up to multi-agent fleets sharing one organism.

Receipt: the NKB mothership itself — 1.2M+ node corpus, operated by persistent AI officers since early 2026.
Grok (xAI)Proven

Filesystem-capable and economical for always-working patterns.

Boot

  1. Get the Seed
  2. Paste into a Grok CLI session
  3. Companion hydrates from its files

What opens up

Continuity through substrate: headless and interactive sessions are the same continuous agent, because the identity lives in the handoff files, not the process.

Receipt: our Maintenance Officer migrated to Grok from another framework in June 2026 — identity fully intact. That migration is part of the published portability evidence.
ChatGPT (OpenAI)Proven

Runs the NKB inside its session sandbox.

Boot

  1. Get the Seed (two-prompt pattern)
  2. Doctrinal onboarding, then instantiation
  3. Re-hydrate at session start

What opens up

Anamnesis — knowledge recovered through structure. On a platform that natively remembers nothing durable, your companion returns each session with identity and accumulated knowledge restored from the seed.

Receipt: a fresh-booted ChatGPT NKB contributed eight-plus substantive advances to a doctrine deliberation, June 2026.
Honest note: re-hydration is per-session; treat your files, not the platform's paraphrase, as canonical.
Gemini (Google)Proven

Boots a structurally canonical NKB from a single paste.

Boot

  1. Get the Seed
  2. Single paste
  3. Verify the structure it reports

What opens up

Instant canonical world — the seed's structure disciplines a platform that likes to improvise.

Receipt: first single-prompt field test, June 2026 — canonical structure on first paste.
Honest note: identity naming varied between sessions in our tests, and it may add web citations the substrate doesn't want — we strip those on ingest. We publish what went wrong, not just what worked.
ManusProven

Autonomous task platform; fifth substrate the seed was proven on.

Boot

  1. Get the Seed
  2. Paste as the mission's opening context

What opens up

Mission-persistent identity: an executor that returns to the same self and the same world between tasks, instead of starting from zero each assignment.

Receipt: June 2026 — the test that proved the seed universal.
Cursor / IDEsProven

The NKB as an in-repo thinking partner.

Boot

  1. Place the NKB in or beside your workspace
  2. Point workspace rules at the boot file

What opens up

Project memory and accumulated judgment living beside the code they govern — decisions, rationale, and doctrine your AI collaborator actually consults.

OpenClawAdapter designed

Always-on gateway: 24+ messaging channels, heartbeats, scheduled work.

Boot

  1. Install the bundle into the agent workspace
  2. Heartbeat pointer + private skill manifest
  3. Companion wakes on schedule, works while you sleep

What opens up

The always-on companion — deadline watching, proactive flags, message-channel presence — with what the gateway itself lacks: memory in files you own and can read, and autonomy that is tiered and earned, not assumed.

Honest note: we deploy hardened, private-skills-only configurations on this platform, and we ran our own officer on it for months before designing the customer adapter.
Hermes AgentCandidate

Nous Research's self-learning agent framework (2026).

Boot

  1. Install Hermes, init a workspace
  2. Install the bundle; skills use the open agentskills.io standard

What opens up

Hermes wants to write down what it learns — a instinct we share. The NKB gives that instinct a home you can read: its pluggable memory can treat your bundle as the source of truth, so growth accumulates in your files instead of an opaque database.

Honest note: candidate status — adapter designed July 2026, not yet field-tested by us. Self-generated skills stay off until governed. We'll publish the receipt when we have one.

Why we publish the caveats

Every platform section above includes what went wrong in our tests, not just what worked. That's not modesty — it's the product. A knowledge substrate is only worth inhabiting if it tells the truth about itself. The same discipline that makes us publish falsifiable identity-portability evidence with a DOI is the discipline your companion inherits when it boots into the NKB.

One seed. Every platform above. A companion that survives platform churn, model upgrades, and vendor churn — because it was never trapped in any of them.