Bilinc

What atakanelik34/Bilinc does

atakanelik34/Bilinc provides a hosted state layer for agents that need more than similarity-based document retrieval. It stores durable memories in Bilinc Cloud and associates writes with their source, such as a run, tool, or operator. The service can expose contradictions as changes to inspect rather than silently replacing previous state.

The system is intended for long-running coding, support, research, and assistant workflows. It can preserve decisions and other agent state, correct existing entries, and remove obsolete state from active recall. It also supports checkpoints for recovering from unsafe or incorrect runs.

How it works

The Python package includes a cloud SDK, command-line interface, and MCP adapter. The adapter communicates over standard input/output and uses an API key saved by the CLI or supplied through the environment. MCP clients call the hosted memory API through tools rather than accessing local storage directly.

Writes return a version that can be retained for optimistic concurrency. A deliberate revision must target an existing memory, which keeps a correction distinguishable from an accidental new write. Snapshot, diff, and rollback operations provide a recovery workflow: create a checkpoint, preview changes, and explicitly confirm a rollback when restoration is required.

Recall accepts a retrieval profile and returns prior context and decisions. The remote MCP connection can also indicate how recently memories were confirmed and flag entries that may be outdated. The hosted package does not include the repository's local runtime internals, storage backends, benchmarks, or operator tooling.

Setup and configuration

Install the package with pip install -U bilinc, then run bilinc login and bilinc quicktest. Browserless environments can use bilinc login --device, while automation can authenticate with bilinc login --api-key bil_live_... or the BILINC_API_KEY environment variable.

After authentication, run bilinc mcp install --client claude-code or bilinc mcp install --client claude-desktop. For other clients, bilinc mcp install prints a generic mcpServers configuration. The adapter uses the saved credential, so the generated configuration normally does not contain an API key.

A remote MCP connection is also available at https://mcp.bilinc.space/mcp. Claude can add it as a custom connector, and Claude Code can register it with an HTTP transport. The remote connection does not require a local Python installation.

Tools and capabilities

The atakanelik34/Bilinc MCP server exposes these core operations through its Python stdio adapter:

• commit_mem: Store durable state with provenance and return a version.
• recall: Retrieve stored context using a selected profile.
• revise: Correct an existing entry without creating a new one.
• forget: Remove active state; a reason is required and the action is audited.
• status: Show workspace, plan, capabilities, limits, profiles, and usage.
• snapshot: Create or list checkpoints.
• diff: Compare a checkpoint with another checkpoint or current state, with values redacted by default.
• rollback: Preview or explicitly execute restoration from a checkpoint.

The remote MCP connector provides ten tools, while the Python stdio adapter provides eight. Local-only claims, contradiction, graph, benchmark, export, import, and replay tooling is not included in the hosted adapter.

Limitations and notes

atakanelik34/Bilinc is cloud-only in the public PyPI package; it does not provide a fully self-hosted runtime. The package is licensed BUSL-1.1, so it is not an Apache-2.0 self-hosted alternative. Bilinc is a memory and state layer, not an agent orchestration framework. For document-only semantic search, a vector database may be a simpler fit.