TechTide Swarm 357

Layered agent orchestration for business automation -- 357 Claude AI agents across 6 business layers, backed by portable Memvid memory and honest cost controls.

                  Management (10) -- Conductor, strategy, memory curation
                       |
    +--------+---------+---------+---------+---------+
    |        |         |         |         |         |
  Sales    Support  Marketing   SEO    Research  Operations
   (62)     (55)      (68)     (47)     (58)       (57)

Install

pip install -e "packages/techtide-swarm[dev]"
swarm demo

swarm demo works with or without an API key -- it shows the architecture and a sample stub when no key is set, or runs a live agent when ANTHROPIC_API_KEY is configured.

What This Does

• swarm run <task> -- Routes a task through the Conductor, which selects agents by role across layers. Passes context sequentially. Shows real cost and latency.
• swarm boot -- Loads the 357-agent roster from config/swarm.yaml, validates soul files, and prints the layer manifest with budget allocations.
• swarm agent --list -- List all agents. swarm agent <name> --run "task" to run a specific one.
• swarm dream -- Runs memory consolidation: loads .swarm/topics/, detects contradictions via string overlap.
• swarm plan <task> -- Deep planning via Claude Opus.
• swarm eval -- Runs 5 benchmark tasks, scores by keyword overlap, compares against baseline.
• swarm status / swarm cost -- Layer health and cost dashboards from local telemetry.

Architecture

Component	Location	What it does
Python package	packages/techtide-swarm/	Agent, Swarm, CostController, MemoryManager, BashSecurityGate, CLI
Memvid bridge (Rust)	packages/memvid-swarm-bridge/	CLI binary for .mv2 create/put/search/verify
Agent roster	config/swarm.yaml	357 agent entries across 7 layers with roles, models, budgets
Soul templates	templates/soul/	8 personality files with YAML front-matter + system prompts
Eval harness	evals/	5-task benchmark with keyword scoring and baseline comparison
Landing page	.ui_landin_sample/minimal/	Next.js 16 template (content in progress)
Memvid core (vendored)	.repos and items/memvid-main/	Upstream Rust library referenced by bridge Cargo.toml

Key Differentiators

Portable .mv2 memory -- Each agent layer gets a single-file Memvid store with WAL-based crash safety, full-text + vector search, and integrity verification. No database server needed.

Business-layer ontology -- Not just "agent A talks to agent B." Six domain layers + management meta-agents model a real organizational structure.

Layered cost controls -- Per-agent budget caps (enforced during Agent.run()), per-layer daily limits, and automatic model downgrade at 80% utilization.

BashSecurityGate -- 13-pattern regex validator wired into the Bash tool. Blocks destructive commands, secret exfiltration, and dangerous system operations. 50+ tests.

Feature Maturity

See STATUS.md for the full breakdown. Summary:

Feature	Status
Agent + Swarm orchestration	Beta
CLI (all 11 commands)	Stable/Beta
Memory (flat-file + Memvid)	Beta
BashSecurityGate	Stable
Cost controls	Beta (enforcement: Alpha)
Dream cycle	Alpha
Eval harness	Alpha
WebSearch tool	Stub

Why Not [X]?

Framework	Swarm 357 advantage	Their advantage
LangGraph	Portable .mv2 memory; business-layer ontology	Durable checkpointing; larger ecosystem; PyPI package
CrewAI	Enforced cost controls; security gate; layered architecture	Faster time-to-first-value; YAML crew config
OpenAI Agents SDK	Multi-agent orchestration; memory persistence	Input/output guardrails; simpler API surface

Memvid Bridge

Build (requires Rust 1.85+):

cd packages/memvid-swarm-bridge && cargo build --release

Set MEMVID_SWARM_BRIDGE to the binary path. See docs/MEMVID_BRIDGE.md.

Development

make install    # pip install editable + dev deps
make test       # pytest (56 tests, <1s)
make lint       # ruff check
make typecheck  # mypy strict
make all        # install + lint + typecheck + test

See CONTRIBUTING.md for the full guide.

CI

GitHub Actions: .github/workflows/swarm357-ci.yml

• Python: install, ruff, mypy, pytest
• Roster: validate 357 agent counts
• Next.js: build + typecheck
• Rust: cargo build + bridge integration tests

Required Environment

ANTHROPIC_API_KEY=sk-ant-...    # for live agent execution

Optional:

SWARM_MODEL_OPUS=claude-opus-4-6
SWARM_MODEL_SONNET=claude-sonnet-4-6
SWARM_MODEL_HAIKU=claude-haiku-4-5-20251001
MEMVID_SWARM_BRIDGE=/path/to/memvid-swarm-bridge

Deployment

Railway

Two Railway services back this product:

Service	Build	Root	Health check
backend	Dockerfile	/	/api/health
frontend	Nixpacks (Next.js 16)	.ui_landin_sample/minimal/	/

# 1. Authenticate & link
railway login
railway init              # creates a new Railway project

# 2. Deploy backend (uses root Dockerfile + railway.toml)
railway up --service backend

# 3. Set backend env vars (replace with real values)
railway variables set ANTHROPIC_API_KEY=sk-ant-... --service backend
railway variables set SUPABASE_URL=https://xxxx.supabase.co --service backend
railway variables set SUPABASE_SERVICE_KEY=eyJ... --service backend

# 4. Deploy frontend
railway up --service frontend --rootDirectory .ui_landin_sample/minimal

# 5. Wire services together
#    - Get backend URL: railway domain --service backend
#    - Set on frontend:
railway variables set NEXT_PUBLIC_API_URL=https://<backend>.up.railway.app --service frontend
#    - Allow frontend origin on backend CORS:
railway variables set ALLOWED_ORIGINS=https://<frontend>.up.railway.app --service backend

# 6. Verify
curl https://<backend>.up.railway.app/api/health   # {"status":"ok","agents":357,...}
curl -I https://<frontend>.up.railway.app/          # HTTP/2 200

Custom domain: Railway dashboard → Service → Settings → Domains → Add Custom Domain, then add a CNAME record at your DNS provider pointing to the Railway-assigned hostname. TLS is auto-provisioned.

License

Apache-2.0