//001
Break the Integration Bottleneck
Models are easy. Connecting them to auth, normalization, and permissions is where progress stalls. We help you skip this phase.
The missing infrastructure layer between AI and enterprise systems
[n.01/05]> About FIM One
Source-available and self-hostable from day one
Single-binary deployment, zero external dependencies
Born from real-world enterprise deployments
Enterprise knowledge lives in dozens of disconnected systems, and the gap is widest where global SaaS meets the China stack. FIM One is the self-hosted all-in-one agent platform that wires every system you already run through one agent core, bridging Global × China so teams can build AI workflows that actually work in production.
[n.02/05]> Our Story
/ Why We Built This /
We kept seeing the same pattern: teams would get a language model working in a demo, then spend months trying to connect it to their actual systems. The model was the easy part. The hard part was everything around it: authentication, data normalization, permission boundaries, and making AI cite its sources so people would actually trust the output.
Every team was rebuilding the same infrastructure from scratch. The same connector code. The same orchestration logic. The same guardrails.
So we built the layer that was missing. FIM One is not another AI framework. It's the production infrastructure that sits between language models and the systems they need to access. Connectors, orchestration, knowledge pipelines, and security boundaries: all in a single self-hostable package that deploys in minutes, not months.
//002
Stop Rebuilding Infrastructure
Stop rebuilding the same connector code and orchestration logic from scratch for every project.
//003
Production-Ready Layer
A production-grade layer for connectors, RAG, and security that deploys in minutes, not months.
[n.03/05]> Principles
/ Our Technology Convictions /
//001
Provider-agnostic: no vendor lock-in
FIM One works with any LLM provider that supports the OpenAI API specification. Switch between OpenAI, Anthropic, DeepSeek, Ollama, or any compatible provider without changing a single line of code. Your AI infrastructure should never be hostage to a single vendor's pricing or availability.
//002
Connector-first: integration is the core value
The bottleneck for enterprise AI isn’t model intelligence. It’s connectivity. Core business data and workflows are locked inside dozens of systems with incompatible APIs. FIM One builds the middleware layer that securely plugs LLM reasoning into these systems. Intelligence without integration is just a chatbot.
//003
Progressive adoption: Standalone to Copilot to Hub
Enterprise AI adoption hits trust barriers at every level, from engineering leadership to end users. FIM One offers a Standalone → Copilot → Hub progression: start small with one team, prove value with minimal risk, then let results drive expansion. The goal is real returns at every step, not a big-bang rollout.
//004
Bounded autonomy: powerful but controlled
AI agents must be powerful enough to be useful, but controlled enough to be trusted. FIM One enforces bounded autonomy through confirmation gates for sensitive operations, re-plan limits to prevent infinite loops, and token budgets to control costs. Autonomy without guardrails is a liability, not a feature.
[n.04/05]> Vision
/ From single chatbot to enterprise AI infrastructure /
// 001
Start by proving value in one specific workflow. A simple document review agent can evolve into an organization-wide compliance platform.
Proof of Value
// 002
Scale from a single use case to enterprise-wide AI orchestration. Enable support bots to become cross-system operations centers.
Scale Orchestration
// 003
Become the full AI infrastructure layer that orchestrates every system. Go from zero to multi-system orchestration in days.
Infrastructure Layer
[n.05/05]> Product Evolution
17 releases in five months: from first commit to production-ready platform, shipping continuously
Ver.
Date
Core Achievement
v0.1
Feb 22, 2026
MVPReAct reasoning engine + DAG Planner for multi-step task orchestration
v0.2
Feb 24, 2026
Multi-Model + MemorySupport for multiple LLM providers with conversation memory and context management
v0.3
Feb 25, 2026
Web Tools + MCPBuilt-in web tools and Model Context Protocol integration for external tool access
v0.4
Feb 25, 2026
Multi-Turn + Agent ManagementMulti-turn conversation support with full agent creation and configuration UI
v0.5
Feb 28, 2026
Full RAG + Grounded GenerationDocument import, hybrid retrieval, and source-cited generation for verifiable AI answers
v0.6
Mar 1, 2026
Connector PlatformOpenAPI import, AI-assisted connector creation, and progressive disclosure for token optimization
v0.7
Mar 6, 2026
Admin Platform + Multi-TenantOrganization management, RBAC, audit logging, and multi-tenant architecture
v0.8
Mar 20, 2026
Workflow Blueprints + Agent IntelligenceWorkflow automation with triggers, agent skill system, evaluation center, and production hardening
v0.8.1
Mar 29, 2026
Solution Templates + Domain Escalation8 vertical solution bundles, per-turn token budget circuit breaker, ReAct cycle detection, specialist domain escalation
v0.8.2
Apr 10, 2026
Intelligent Document ProcessingAdaptive vision handling, multi-turn vision persistence, smart PDF text/image separation, pre-built data-science sandbox image
v0.8.3
Apr 16, 2026
Document OCR + Turn ProfilerBuilt-in markdown conversion for 15+ formats, vision-aware RAG ingestion, per-user rate limiting, turn-level phase timings
v0.8.4
Apr 17, 2026
Conversation Recovery + Cache ObservabilitySynthetic tool_result persistence for session recovery, SSE stream auto-reconnect, prompt cache token metrics
v0.8.5
Apr 23, 2026
Xinchuang DB + Feishu + Agent HooksKingbaseES / HighGo / DM8 support, Feishu channel IM approval gates, Agent Hook System, Exa deep-reasoning search integration
v0.8.6
May 8, 2026
Billing & SubscriptionsStripe Pro subscriptions, admin-controlled billing module, per-user quotas with unlimited allowlist, full API Reference and Channels docs in every locale
v0.8.7
Jun 10, 2026
Content Guardrails + Patch ToolInput and output content guardrails, file_ops.apply_patch for V4A-format diffs, fixes for a two-factor bypass and OAuth auto-linking on unverified email
v0.8.8
Jun 22, 2026
SSRF Hardening + Extended ThinkingBlocked SSRF bypass via IPv4-mapped IPv6, MCP server URLs now SSRF-validated, overridable Compose database credentials, extended thinking on newer Anthropic models
v0.8.9
Jul 8, 2026
Optional Modules + Resumable RunsSkills and Workflows became opt-in modules, DAG replans reuse completed steps and interrupted runs resume, slow tools run in the background, full transcript kept before compaction
// [#connect]// [#orchestrate]// [#review]// [#connect]// [#orchestrate]// [#review]// [#connect]// [#orchestrate]// [#review]
// [#connect]// [#orchestrate]// [#review]// [#connect]// [#orchestrate]// [#review]// [#connect]// [#orchestrate]// [#review]
Two ways to run