FASS is a runtime platform that securely executes and manages autonomous agents at scale. Click any card below or use the left nav to jump to any section.
You do not need to write code to use FASS. Think of it as a personal AI team — agents that handle tasks for you, supervised by a system that keeps everything safe and auditable.
Organize Your Files
Connect Google Drive. FASS automatically discovers, classifies, and organizes your files using AI — no manual tagging needed.
01Go to Ingest Pipeline in the sidebar
02Click Connect Google Drive
03Click Run Discovery — FASS does the rest
Talk to Francis
Francis is your Telegram-based AI assistant. Send a message, get things done — from scheduling to research to running FASS commands.
01Open Telegram and search @FassBot
02Send /start and link your FASS account
03Type any task in plain English
Automate Your Workflow
Describe what you want to automate in plain English. FASS translates it into an agent task with a scoped authority token — no code required.
01Go to the Overview dashboard
02Click New Task and describe your goal
03FASS assigns the right agent automatically
Monitor Everything
Every action your agents take is logged in plain language in the Audit Log. You always know what happened, when, and why.
01Click Audit Log in the sidebar
02Filter by agent, date, or event type
03Export to CSV for records
Built-in AI
Meet Francis & Forge
FASS ships with two native AI personalities. Both are model-agnostic — you choose which LLM powers them. They operate under the same authority envelope system as all other agents.
🤝
Francis
Telegram Work Bot · Personal Assistant
Francis lives in Telegram. He is your always-on work companion — communicate with your projects, run FASS commands, get summaries, and delegate tasks in plain English from your phone. Designed for normal users: no code, no dashboards, just conversation.
→Runs on Telegram — zero setup for end users
→Executes FASS commands on your behalf
→Summarizes audit logs and agent activity
→Answers questions about your project in natural language
→Model-agnostic: OpenAI, Gemini, Claude, or any OpenAI-compatible API
⚙️
Forge
Manus Architecture Bot · Dev Scaling Agent
Forge is the developer power tool. Built on the Manus architecture, Forge handles complex engineering tasks: scaffolding projects, auditing codebases, managing deployments, and scaling infrastructure. Designed for developers who want a senior engineer on demand.
→Scaffolds full-stack projects from a single prompt
→Runs fass audit + repair autonomously
→Manages Vercel and Cloudflare deployments
→Monitors and alerts on system health
→Model-agnostic: swap the underlying LLM at any time
Model Configuration
Choose Your LLM
FASS is fully model-agnostic. Click any model below to assign it to an agent. Every model runs inside an authority envelope — the agent permissions are independent of which model powers it.
GPT-4.1
OpenAI
Pro
Gemini 2.5 Flash
Google
Free
Claude 3.5 Sonnet
Anthropic
Pro
Llama 3.3 70B
Meta (self-host)
Enterprise
Mistral Large
Mistral AI
Pro
DeepSeek R1
DeepSeek
Free
Assign GPT-4.1 to an agent
bash
fass config set --agent francis --model "gpt-4.1"
Developer
CLI Reference
Every fass command with flags and copy-ready examples. Install: curl -fsSL https://fass.systems/install.sh | bash
Every agent action in FASS is wrapped in an authority envelope — a cryptographically signed token that defines exactly what the agent is permitted to do, on what resource, and for how long. No envelope means no execution. Enforced at the perimeter.
Field
Type
Description
agent_id
string
The unique identifier of the agent this token is issued to
scope
string[]
Array of permitted actions: read, write, execute, witness
resource
string
The specific resource or namespace the agent may access
ttl
number
Token lifetime in seconds. Default 3600. Max 86400.
issued_at
timestamp
Unix timestamp of token issuance
expires_at
timestamp
Unix timestamp of token expiry
signature
string
RSA-SHA256 signature over the token payload
witness_hash
string
SHA-256 hash of the issuance event, stored in audit log
FASS agents do not receive orders — they make promises. Based on Promise Theory (Burgess and Bergstra), each agent independently declares what it can do and commits to doing it. This eliminates central coordination bottlenecks and makes the system resilient by design.
🎯
Step 01
Objective Declared
The user or orchestrator declares an objective — a high-level goal with no prescribed execution path.
🤝
Step 02
Agents Make Promises
Each agent independently evaluates the objective and makes a promise: 'I can handle this subtask.' No central coordinator assigns work.
🔐
Step 03
Authority Envelope Issued
The Authority System issues a signed token scoped to the specific task. The agent cannot act outside its envelope.
⚡
Step 04
Independent Execution
Agents execute their promised subtasks in parallel, writing results to the shared state store. They never block each other.
📋
Step 05
Witness Logged
Every significant action is hashed and logged to the immutable audit trail. The system can reconstruct exactly what happened and when.
✅
Step 06
Objective Fulfilled
The orchestrator aggregates results. If any agent fails its promise, the system retries or reassigns — without human intervention.
Infrastructure
System Architecture
FASS separates the control plane from the data plane. The control plane handles auth, config, and policy. The data plane handles all execution. This prevents the control layer from becoming a bottleneck at scale.
Control Plane
→Next.js UI (Vercel Edge)
→REST API (20+ routes)
→Agent Registry
→Authority System
→Audit Log (Supabase)
Data Plane
→Agent Workers (Cloudflare)
→Task Queue (Redis Streams)
→Execution Runtime
→Ingest Pipeline
→Witness Logger
Gateway Layer
→fass-root Worker (CF)
→fass.systems routing
→www to apex redirect
→/install.sh endpoint
→/health endpoint
Storage Layer
→PostgreSQL (Supabase)
→Redis (task queue)
→Object Storage (R2)
→KV Store (CF KV)
→Audit log (immutable)
Infrastructure
Deployment
FASS deploys to Vercel (control plane) and Cloudflare Workers (gateway). Both are triggered automatically on push to main.
Deploy to Vercel
bash
fass deploy --target vercel
# or: git push origin main (auto-deploys via GitHub)
Deploy a Cloudflare Worker
bash
cd your-worker-dir
npx wrangler deploy
# Routes defined in wrangler.toml