// CLOUD SDK · 7-DAY TRIAL · MCP READY

BIL.INC

MEMORY THAT AGENTS CAN VERIFY, REVISE, AND ROLL BACK

Bilinc 2.0 starts with a cloud-only SDK and a 7-day hosted trial. Install the SDK, create a hosted API key, then commit and recall memory from your agent runtime.

pip install bilinc
Python 3.10+ · BUSL-1.1
START 7-DAY CLOUD TRIALINSTALL CLOUD SDKADD MCPSOURCE ACCESS
5k+ PyPI downloadsPyPI v2.0.0MCP ready7-day trial
[ v2.0.0 STABLE ]█
20
MCP tools
98.0%
LongMemEval-s R@5
7 days
Cloud trial
Cloud API
Storage path
Live
Cloud billing

what.is.bilinc

Bilinc is a memory and state layer for AI agents.

It lets agents save important facts, recall prior decisions, detect contradictions, and roll back bad state. Start with the Python SDK or MCP adapter against Bilinc Cloud; the 2.0 package is cloud-only and keeps local storage internals out of the public install path.

Install BilincAdd MCPView use cases

Verified writes

Run checks before agent state becomes durable memory.

Conflict-aware memory

Revise or contract beliefs when new state contradicts old state.

Snapshots and rollback

Checkpoint, diff, and restore agent memory after a bad run.

MCP for coding agents

Attach Bilinc to Codex, Claude Code, Cursor, and local agent workflows.

Cloud API keys

Use the PyPI SDK with hosted API keys instead of shipping local storage internals.

Cloud for teams

Use hosted API keys, shared workspaces, and billing-backed entitlements.

why.not.just.vector.memory

Vector databases retrieve similar text. Bilinc manages agent state.

Long-running agents do not only forget. They also remember the wrong things, overwrite useful context, and lose the trail behind important decisions. Bilinc wraps memory with the controls needed for durable state.

Compare memory modelsReview boundaries
# production memory questions
  • Was this memory verified before it was written?
  • Did it contradict prior agent state?
  • Which run, tool, or operator created it?
  • Can the team audit why state changed?
  • Can the agent roll back after a bad write?

Vector memory

Retrieves similar text and helps with semantic search.

Bilinc state plane

Adds verification, revision, provenance, snapshots, rollback, and audit around memory.

operator routes

Pick the path that matches your deployment reality.

Bilinc 2.0 starts with a cloud-only SDK and a 7-day hosted trial. Paid Cloud exists for teams that need shared state, retention, and production accountability.

start here

Connect your agent

Install the cloud-only SDK, start a 7-day trial, create a hosted API key, and use commit, recall, status, and MCP from your agent runtime.

Install SDK

hosted path

Evaluate Bilinc Cloud

Start a 7-day hosted trial, test managed state and API-key workflows, then move to Pro or Team from the billing console when the workload is real.

Start trial

operational trust

Check the control surface

Read how writes, rollback, observability, audit trail, and runtime entitlements behave before you put persistent agents on top.

Review trust

billing posture

Understand billing status

Cloud billing is live for owner/admin operators. Internal entitlements remain the runtime authority; Stripe syncs subscription state.

Read billing

// MODULE_LIST.EXE

CORE CAPABILITIES

Everything a production state layer for autonomous agents should have.

01. VERIFIABLE
_
□
×

VERIFIABLE STATE

Z3-backed verification before every commit. State passes validation logic — never blindly stored. Full audit trail for replayable history.

PRODUCTION-CAPABLE
02. BELIEF
_
□
×

BELIEF REVISION

AGM-style belief revision machinery for changing or conflicting agent state. Controlled updates under contradiction — no silent data loss.

SUPPORTED
03. DURABLE
_
□
×

DURABLE PERSISTENCE

Hosted storage and entitlement gates. Snapshot, recall, and audit primitives for production durable state workflows.

PRODUCTION-CAPABLE
04. OBSERVABILITY
_
□
×

OBSERVABILITY

Real operator surface. /health (liveness + readiness) and /metrics (Prometheus-compatible) over authenticated HTTP. bilinc_ prefix metrics.

PRODUCTION-CAPABLE
05. KNOWLEDGE
_
□
×

KNOWLEDGE GRAPH

Semantic memory projected into a traversable graph. Contradiction analysis, structural queries, and graph-based belief coherence checks.

SUPPORTED
06. MCP
_
□
×

MCP + AUTH

20 MCP tools over stdio (trusted local) and HTTP (bearer auth + rate limiting). Production-hardened transport with 401/429 enforcement.

PRODUCTION-CAPABLE

// BOOT_SEQUENCE.SH

HOW IT WORKS

Five commands to a trustworthy agent state layer.

BILINC_TERMINAL — bash
BILINC SDK v2.0 • BUSL-1.1 • source access by request

use.with.your.agent

Start with one hosted memory contract, then wire it into your agent.

Bilinc should be easy to test from a single terminal and still make sense when the same agent workflow becomes a team-owned production surface.

Codex / Claude / Cursor

Paste the MCP config and give your agent a memory contract before risky work.

Agent recipes

Python agents

Use CloudClient directly for hosted commit, recall, status, and durable memory operations.

Python SDK

Hosted Cloud API

Issue workspace-scoped API keys and call commit/recall over authenticated HTTP.

Cloud quickstart

Team operations

Use shared workspaces, billing entitlements, and audit history when agent state becomes operational.

Billing posture
Recommended first run:pip install bilinc->/docs/mcp->/docs/agent-recipes

// BENCHMARK_RESULTS.LOG

BENCHMARK RESULTS

Reproducible benchmark receipts. No LLM in the loop; not hosted-SLA claims.

LongMemEval-s R@5: 98.0%ConvoMem: 98.0%LoCoMo: 90.3%
01. LONGMEMEVAL
_
□
×
98.0%

Long-horizon single-session memory recall. No language model in the loop — measured pipeline.

NO LLM · REPRODUCIBLE
02. CONVOMEM
_
□
×
98.0%

Conversational memory across facts, preferences, events, instructions, and entities.

5 CATEGORIES · REAL PIPELINE
03. LOCOMO
_
□
×
90.3%

Temporal reasoning, causal chains, and multi-hop retrieval under long-context pressure.

TEMPORAL · CAUSAL · MULTI-HOP
SCORE_PROGRESSION — BASELINE → BILINC
_
□
×
CONVOMEM
17.5%
98.0%
LOCOMO
9.1%
90.3%

// ARCH_MAP.SYS

SEVEN-LAYER ARCHITECTURE

One coherent state plane. Every layer with a clear responsibility.

L1
CLOUD SDK LAYER
CloudClient — commit, recall, status via hosted API
L2
HOSTED PERSISTENCE
Bilinc Cloud — durable storage, retention, entitlement gates
L3
VERIFICATION & AUDIT
Z3-backed verification — replayable state history
L4
BELIEF MANAGEMENT
AGM revision — controlled updates under contradiction
L5
KNOWLEDGE GRAPH
Semantic memory — traversal + contradiction analysis
L6
TRANSPORT LAYER
MCP stdio (local) + HTTP (bearer auth + rate limiting)
L7
OPERATIONS LAYER
/health + /metrics — Prometheus-compatible monitoring

// SYSTEM_ROADMAP.DAT

DEVELOPMENT ROADMAP

Where we are. Where we're going.

PHASE 01 — CORE STATE ENGINE
_
□
×
v1.0.4 — SHIPPED
  • ✓CloudClient API (commit, recall, status)
  • ✓Hosted persistence + entitlement gates
  • ✓Snapshot, diff, rollback primitives
  • ✓Z3-backed verification layer
  • ✓AGM belief revision engine
  • ✓Knowledge graph module
PHASE 02 — PRODUCTION TRANSPORT
_
□
×
v1.0.4 — SHIPPED
  • ✓MCP stdio transport (trusted local)
  • ✓MCP HTTP (bearer auth + rate limit)
  • ✓20 MCP tools surface
  • ✓/health + /metrics (Prometheus)
  • ✓Cloud runtime and billing guardrails
PHASE 03 — EVIDENCE-AWARE RECALL
_
□
×
MAIN — LANDED
  • ✓Recall capture and replay metrics
  • ✓Attributed claim projection
  • ✓Read-only contradiction probes
  • ✓Recall profiles: fast / balanced / verified / deep
  • ✓Entity and backlink projection
PHASE 04 — INTEGRATIONS
_
□
×
IN PROGRESS
  • ►LangGraph official adapter
  • ►CrewAI plugin
  • ►AutoGen bridge
  • ►OpenAI Agents SDK hook
  • ►Claude / Anthropic SDK integration
PHASE 05 — HOSTED CLOUD
_
□
×
LIVE CLOUD
  • ►Public pricing published
  • ►Entitlements remain runtime authority
  • ►Pro and Team checkout live
  • ►Managed retention and team operations
  • ►Scale and enterprise rollout review

faq

Practical answers before you wire memory into an agent.

Is Bilinc a vector database?

No. Bilinc can use retrieval signals, but its product boundary is agent state: typed memory, verification, revision, audit history, snapshots, and rollback.

Is the PyPI package self-hosted?

No. Bilinc 2.0 on PyPI is cloud-only: SDK, CLI, and MCP adapter for hosted Bilinc Cloud.

Do I need a Cloud key?

Yes for Bilinc 2.0. Start a 7-day trial, create an API key, then use PyPI, CLI, or MCP from your agent runtime.

What happens when memory contradicts older state?

Bilinc exposes conflict-aware revision paths so a new write does not silently overwrite useful state. Operators can inspect the resulting history.

Does Bilinc make every answer true?

No. Verification improves inspectability and failure handling. It does not turn every generated answer into automatic truth.

Which agents can use it?

Any agent that can call Python, HTTP, or MCP can use Bilinc. The docs include Codex, Claude Code, Cursor, and general agent prompts.