Every AI
operation.
Verified.
Rezos wraps any LLM with cryptographic verification. Every file operation your agents claim to perform is validated against the actual filesystem — before you ever see the result.
Strict Write Discipline
Pre/post SHA-256 snapshots verify every file operation an AI agent claims. If reality doesn't match the claim, Rezos flags it and triggers a correction turn — automatically.
[SWD] Snapshot: src/auth.ts [SWD] sha256:a3f9c2b1... ← pre [SWD] Writing 247 bytes... [SWD] sha256:b7e4d182... ← post [SWD] ✓ Verified
Multi-provider routing
Route between Anthropic, OpenAI, and DeepSeek with automatic circuit breakers. Bring your own keys. Cost-aware fallbacks switch providers when one fails or exceeds budget.
Provider: Anthropic (primary) Circuit: CLOSED Fallback: OpenAI → DeepSeek Budget: 84k / 500k tokens
Self-healing memory
A SQLite FTS5 index rebuilt from MEMORY.md as the authoritative source. Survives crashes, compresses automatically past 100 entries, and stays accurate across sessions.
$ rezos dream Compressing 118 entries... ✓ Rebuilt index (34 entries) ✓ MEMORY.md updated
SWD Receipts & audit trail
Every verified session produces a tamper-evident receipt in JSON. Contains touched files, pre/post hashes, provider identity, cost, and verification result. Built for teams that need to prove what changed.
{
"session": "#4831",
"verified": true,
"drift": 0,
"cost_usd": 0.06,
"files": [...]
}MCP Adapter
Expose SWD tools over Model Context Protocol via stdio transport. Any MCP-compatible agent — Claude Code, Cursor, your own framework — can route file operations through Rezos's verification layer.
// MCP host config
{
"servers": {
"rezos": {
"command": "rezos mcp"
}
}
}Skill Packs
SKILL.md files (project-local or user-global) encode rules for the model — your conventions, forbidden patterns, architecture decisions. Reviewable, versioned, composable.
$ rezos learn --depth deep Analyzing 247 files... ✓ SKILL.md generated 18 conventions captured
Verification that pays
to hold the token.
$REZOS is not a governance token that sits idle. Every session that runs through the verification layer consumes tokens — burned forever. Teams that stake reduce their costs. Growth compounds scarcity.
Usage → Deflation
Every verified operation consumes $REZOS proportional to the number of file actions in the session. Tokens are sent to a dead address — permanently removed from supply. More teams shipping with AI means more burns.
Lock → Discount
Teams stake $REZOS to reduce their per-verification cost. The more staked, the lower the fee. Staked tokens are locked for a minimum of 30 days, reducing circulating supply and aligning team incentives with the protocol.
Hold → Vote
Token holders vote on protocol parameters: burn rate adjustments, supported LLM providers, verification fee schedules, and treasury allocations. Weight is proportional to staked balance — not raw holdings.
Full CLI, no account required.
- Strict Write Discipline (SWD)
- Multi-provider BYOK routing
- SQLite memory & FTS5 index
- SWD receipts & audit trail
- MCP adapter
- Skill Packs
- Drift detection
- Token budgeting
Dashboard, shared receipts, analytics.
- Everything in Open Source
- Web dashboard & live feed
- Shared team receipts
- Cost analytics per agent
- Slack & PagerDuty alerts
- SSO / SAML
- Custom skill registry
- Priority support (< 4h SLA)
On-prem, SOC 2, dedicated success.
- Everything in Teams
- On-premises deployment
- SOC 2 Type II
- Custom data retention
- SLA guarantees
- Dedicated success engineer
- Custom integrations
- Volume pricing
— Stop trusting. Start verifying.
Your first session.
Fully verified.
Install in 30 seconds. Your first rezos chat comes with full SWD verification, budget tracking, and a tamper-evident receipt — no configuration required.