Buying snapshot
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Decision point | What LedgerMind states |
|---|---|
Licence | Mixed by component: the public repository, Integrations and Protocol are Apache-2.0; Local uses Business Source License 1.1; Core is proprietary |
Commercial and managed options | A commercial licence for production uses outside Local's Additional Use Grant; LedgerMind Cloud, a managed API; enterprise evaluation, deployment and licensing through LedgerMind |
Installation | Self-hosted 4.1.0 package for Linux x86_64; installs without root access; Docker is not required |
What you supply | A supported agent, a generation provider and model with strict JSON Schema structured outputs, and an OpenAI-compatible embedding API or the supported local embedding runtime |
Which path fits
Internal tool or non-commercial use. Local's Business Source License permits internal-tool and non-commercial deployments under its Additional Use Grant. Read the licence file shipped with the release you install; LedgerMind says that file is authoritative.
Production use outside the grant. This requires a commercial licence. LedgerMind directs platform and enterprise deployments to contact it for evaluation, deployment and licensing.
You would rather not run it. LedgerMind Cloud is the managed API: the user plugin connects the agent with a workspace API key, and memory processing and storage run in the managed service. Accounts and workspaces are at ledgermind.org.
What each component is and its licence
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Component | Purpose | Source and licence |
|---|---|---|
Integrations | Collects completed agent interactions and delivers them as versioned RawRounds | Public source, Apache-2.0 |
Protocol | Public contracts, schemas and conformance fixtures | Public source, Apache-2.0 |
Local | Installer, local service, provider profiles, secrets, retention, backup and egress audit | Source available, Business Source License 1.1; changes to GPLv3-or-later on the stated Change Date |
Core | Durable knowledge formation, revision and retrieval | Proprietary signed local binary |
The Apache-2.0 licence on the public repository does not cover Local or Core. LedgerMind asks you to consult the licence included with each component before use or redistribution.
How the cost builds up
On the self-hosted path, the licence is only one part of the bill. You also pay for the models LedgerMind calls and the Linux host it runs on. The operator chooses every model and endpoint. With local generation and embedding models, the complete memory workflow can stay on your machine. With a remote provider, the needed model payload leaves through Local to the endpoint you selected.
The generation model does semantic extraction and resolution. For reliable production behaviour, LedgerMind recommends a model with 120B parameters or more on an endpoint that supports strict JSON Schema structured outputs. Its tested reference model is deepseek/deepseek-v4-flash-0731; another model may be used when it meets the same size and structured-output requirements.
What drives the bill
Generation provider. Whether you call a remote OpenAI-compatible API or run a model locally decides where model spend lands. With OpenRouter, you pick one primary provider route and, optionally, one fallback.
Embeddings and reranking. Embeddings use an API or the signed local CPU runtime. Reranking is optional: disabled, a Cohere-compatible or NVIDIA NIM endpoint, or a separately downloaded local CPU runtime.
Model residency. Local embedding and reranker models can stay loaded for each agent turn (Session), stay warm for a selected period (Idle timeout), or stay resident (Always on), which LedgerMind ties to the lowest recall latency. LedgerMind recommends Session mode.
Agent context. In LedgerMind's own 19 September 2026 benchmark, LedgerMind used 232,429 agent tokens across 20 tasks, against 308,131 for budget-limited history. LedgerMind calls this a development run and says a full rerun on the release binary is still required.
What to check before you commit
Platform limits. The 4.1.0 package supports Linux x86_64 with glibc 2.31 or newer and needs bubblewrap; no aarch64 package is included. Docker is not a supported secure deployment path for 4.1.0.
Backups and retention. A normal uninstall keeps memory and creates a local backup when data exists. Such backups can contain sensitive workflow data and must be protected like credentials. Raw workflow payloads expire after 30 days by default; the period is configurable.
Our take
If your use fits Local's internal-tool or non-commercial grant, the model calls and the Linux host are the costs you can size yourself; Core is proprietary, so read its licence in the release too. If your use falls outside the grant, treat LedgerMind as a commercial purchase before you build on it.