Skip to content
Comparison · Supervised autonomy

Autonomy you can
supervise — not walk away from.

Zero-human orchestrators optimize for agents that run unattended. FlukeBase optimizes for agents you can trust in production — every destructive op behind a recorded approval, every decision in an auditable trail.

Approval gates · Versioned memory · One binary · 8210 memories indexed · 28121 sessions captured · 859 MCP tools

See the comparison Get started

Approval gates

Destructive ops need a recorded decision

Versioned memory

Every fact has history & a diff

Audit trail

Every action, searchable

One binary

git · CI · deploy · payments

Feature comparison

What "supervised" actually buys you.

The same capabilities, seen through two different priorities: run unattended, or run accountable.

FlukeBase
Zero-human orchestrators
Persistent, bitemporal memory. A versioned knowledge graph with semantic search — every fact carries history, a diff, and the ability to be corrected or forgotten.
Memory optimized for the agent's next step. Context is retained to keep the loop running, not to give a human a durable, reviewable record.
Integrated CI & deploy. Self-hosted git, pipelines, push-to-deploy, and health checks live in the same binary — one dashboard, one audit surface.
CI & deploy live in external tools. The orchestrator drives them, but the pipeline and its history sit outside the agent's own system of record.
Native Lightning payments. Bitcoin sats + USD-pegged invoices via Blink, gated by the same budget and approval controls as every other action.
Payments are out of scope. Money movement, if any, is delegated to a separate service — outside the autonomy engine's guardrails.
First-class audit trail. Every tool call, approval decision, and dollar is recorded, searchable, and exportable across 28121+ sessions.
Logs, not an audit trail. Runs are observable while they execute, but reconstructing "who approved what, and why" after the fact is a manual exercise.
Multi-project operations. One instance runs a whole portfolio — Go, Rails, Next.js, Elixir — with cross-project memory so agents learn once and apply everywhere.
Usually scoped to one task or repo. Coordinating many projects means running many instances, each with its own isolated context.

Different tools for different jobs. If you want an agent that never asks, a zero-human orchestrator fits. If you want one you can put in front of a customer's back office, that's FlukeBase.

Architecture

One binary vs an eight-tool stitch.

A stitched stack is eight auth models, eight billing relationships, and eight places for context to leak. FlukeBase folds them into one.

FlukeBase — one Go binary

A single flukebase process. One auth model, one audit log, one thing to deploy.

  • ✓ Self-hosted git
  • ✓ CI pipelines
  • ✓ Deploy & health checks
  • ✓ Persistent memory
  • ✓ MCP server (859 tools)
  • ✓ Email (send + inbound)
  • ✓ Lightning payments
  • ✓ Approval gates & audit trail

1 auth model · 1 audit log · 1 deploy

A stitched-together stack

An orchestrator plus the services it has to reach for. Each is a separate account, SDK, and bill.

  • + Git host
  • + CI provider
  • + Deploy platform
  • + Vector DB / memory store
  • + Agent orchestrator
  • + Email service
  • + Payments processor
  • + Logging / audit tooling

8 auth models · 8 bills · context scattered across all of them

Give your agents autonomy — and keep the controls.

Free for individual developers. Connect via MCP in seconds. Install the plugin and every session gets memory, 859 tools, and approval-gated ops.

Create free account Connect your agent

Install the FlukeBase plugin

$ claude plugin install flukebase@flukebase click to copy

Then run /flukebase-setup once — it wires the MCP server and authenticates via OAuth.