SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentsBilinc

Bilinc

Hosted agent memory with an audit trail: writes carry provenance, contradictions surface as inspectable changes, and bad runs are recovered from snapshots.

Open website

About this tool

Bilinc is the state layer between an AI agent and the things it must remember. It makes memory writes attributable, correctable, and recoverable, rather than treating retrieval as a pile of similar text. Long-running agents, besides answering "what is similar to this," also need to answer "who wrote this state, was it verified, does it contradict what we already know, can it be undone"—this is the layer Bilinc provides.

Main capabilities:

  • Commit, recall, correct, snapshot, and roll back agent state, while preserving provenance.
  • Every write is built around verification, provenance tracking, and recovery mechanisms.
  • Access hosted memory infrastructure through a single API key.
  • Provides a Python SDK, CLI, and MCP server/adapter.
  • The MCP remote connector includes ten tools; the Python stdio adapter includes eleven tools, namely the eight-step memory lifecycle plus three review tools.
  • "Recall" in the remote connector shows how long it has been since each memory was last confirmed, and marks entries beyond the review window (90 days for facts, decisions, how-to guides, and places; 1 day for working notes; events are never marked) as possibly stale. Content is not hidden or deleted. Correcting a memory with the same value confirms it.
  • A read-only connection lists only four read tools.

How to use:

  • Fastest path: pip install -U bilinc, bilinc login, then run bilinc quicktest against Bilinc Cloud.
  • Claude connector (no Python installation required): In Claude, open Customize → Connectors → Add custom connector, name it Bilinc, paste https://mcp.bilinc.space/mcp。登录并选择只读或读写访问。设置指南见 bilinc.space/mcp.
  • Claude Code remote connector: run /mcp and complete authentication.
  • bilinc login opens a browser. Sign in with Google, GitHub, or email (free, no credit card required) and approve this computer; the CLI saves its own API key with owner-only permissions. The key does not appear in the URL or terminal.
  • Browserless environments (SSH, servers, containers): bilinc login --device prints a short code that can be approved in any browser.
  • CI and scripts: bilinc login --api-key bil_live_..., or set BILINC_API_KEY.
  • Bilinc exposes a standard Model Context Protocol server over stdio, and any MCP-compatible client (Claude Code, Codex, Cursor, Hermes-Agent, opencode, etc.) can connect. After bilinc login, the CLI can write the configuration for the client; command is the interpreter where Bilinc is installed, and the adapter reads the key saved by bilinc login, so the configuration contains no key. If there is no saved key on the machine, you can add "env": { "BILINC_API_KEY": "bil_live_..." }.
  • Documented client setup: Claude Code, Codex, Cursor, any MCP client.
  • Reviewing stored content: listing, history, and export are reads and are not billed; confirm consumes one write. Forgetting a memory also removes its value from history: earlier values of a forgotten memory will not be returned.
  • Rollback has two phases: execute uses the token returned by preview, and never prompts interactively, so it can be used safely in automation.

Use cases (official "When to use Bilinc"):

  • Long-running agents—coding, support, research, or personal assistants—need to recall prior decisions before risky operations.
  • You need to know which run, tool, or operator produced a given agent state.
  • A bad agent run wrote incorrect state, and you need a recovery path rather than manual cleanup.
  • Multiple agents or team members share one memory surface, and you need denKey-scoped access and usage visibility.

Not applicable scenarios (official "When not to use Bilinc"):

  • You only need semantic search over documents—a vector database is a simpler primitive.
  • You require an Apache-2.0 licensed, fully self-hosted runtime. The public package is cloud-only, licensed under BUSL-1.1.
  • You want the memory layer to also serve as an agent framework. Bilinc is a state layer invoked by the runtime; it does not orchestrate agents.

Hosted endpoints:

  • All hosted endpoints are under https://bilinc.space . Authenticated memory operations require a valid Bilinc Cloud entitlement.
  • Send an Idempotency-Key header for any write that may be retried: the same key with the same payload replays the original result and is billed only once; the same key with a different payload is rejected with 409 idempotency_conflict.

Benchmark receipts:

  • Frozen regression receipt, LongMemEval-s cleaned retrieval fixture, 500 questions: Hit@5 98.0%, NDCG@5 0.913, no LLM reranking, no paid APIs.
  • This is a frozen isolated retrieval guardrail—not a current hosted SLA, not an end-to-end agent score, and not a competitor ranking. Published memory system scores use different metrics, datasets, and degrees of LLM assistance, so they are not directly comparable.
  • The repository keeps dated manifests containing source state, dataset provenance, runner, and metric semantics. These are traceability artifacts and do not claim that Bilinc is universally first.

Limitations and notes:

  • The public package is intentionally smaller than this repository: it does not bundle the internal StatePlane or the local storage runtime.
  • Operations and debugging tools—health probes, benchmarks, bulk import, workspace replay—are local-only, as are the cognitive read tools for claims, contradictions, and graph queries. Hosted adapters do not bundle local runtime internals.
  • The official documentation also mentions entries such as the technical article "Why vector search is not enough for agent memory," Bilinc vs. vector memory, Bilinc vs. Letta, what is AI agent memory, how to roll back agent memory writes, what is an MCP memory server, and how to audit what an agent remembers, but the main text does not elaborate on the specific content, so it is not provided here for now.
  • The official links provided include the website https://bilinc.space , sign-up https://bilinc.space/signup , installation guide https://bilinc.space/install , pricing https://bilinc.space/pricing (permanent free tier, no credit card required), quick start https://bilinc.space/docs/quickstart , cloud quick start https://bilinc.space/docs/cloud-quickstart , migration guide https://bilinc.space/docs/migration-v2 , MCP setup https://bilinc.space/docs/mcp 、PyPI https://pypi.org/project/bilinc/ , machine-readable index https://bilinc.space/llms.txt and https://bilinc.space/ai-index.json .
  • The license is BUSL-1.1, see LICENSE for details. Security reports should follow SECURITY.md. Do not include private memory values, API keys, or production logs in issues, pull requests, benchmark fixtures, or screenshots.

Related tools

Browse category