Skip to main content

Run Manasvi Locally

Once you've run pnpm manasvi init, starting everything is a single command.

Start all services​

pnpm manasvi start

The CLI starts all nine services in dependency order, waits for each one to pass its health check, and prints a status table when everything is ready:

  Manasvi  start

  Starting services
  ✔ policy-service        ready  (312ms)
  ✔ approval-service      ready  (284ms)
  ✔ memory-service        ready  (341ms)
  ✔ audit-service         ready  (298ms)
  ✔ execution-manager     ready  (356ms)
  ✔ node-manager          ready  (270ms)
  ✔ orchestrator-service  ready  (489ms)
  ✔ ingress-service       ready  (301ms)
  ✔ api-gateway           ready  (278ms)

  ✔ All 9 services healthy

  Next steps:
  → Chat via terminal:  pnpm cli
  → Check status:       pnpm manasvi status
  → API endpoint:       http://localhost:4100/test-harness/chat

Service logs are written to ~/.manasvi/logs/<service-name>.log.

Check status​

At any time, run:

pnpm manasvi status

This shows each service's health, latency, and your current configuration:

  Services
  API Gateway           :4100   ● healthy (12ms)
  Ingress Service       :4101   ● healthy (9ms)
  Orchestrator          :4102   ● healthy (14ms)
  Policy Service        :4103   ● healthy (8ms)
  Execution Manager     :4104   ● healthy (11ms)
  Memory Service        :4105   ● healthy (10ms)
  Node Manager          :4106   ● healthy (9ms)
  Audit Service         :4107   ● healthy (8ms)
  Approval Service      :4108   ● healthy (9ms)

  ✔ All 9 services healthy

  Configuration
  Profile    local
  Model      Mock (testing mode)
  Channels   none
  Docs UI    http://localhost:3000

Send a test message​

Option A — Interactive terminal​

pnpm cli

This opens an interactive REPL that sends messages through the API gateway. Type a message and press Enter to chat with the agent.

Manasvi terminal  (session: session:abc123)
Type a message, or /help for commands.

You: What can you help me with?
Agent: I can help you with...

Option B — Direct API call​

curl -X POST http://localhost:4100/test-harness/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello, what tools do you have?", "actor": "user:test"}'

Stop services​

pnpm manasvi stop

Sends SIGTERM to each running service and waits up to 5 seconds for it to exit cleanly. If a service doesn't stop in time, it prints an error and exits with a non-zero code.

If services are hung and won't stop gracefully, use --force to send SIGKILL:

pnpm manasvi stop --force

To restart all services:

pnpm manasvi restart

# Or force-kill stuck services then restart:
pnpm manasvi restart --force

Service ports reference​

ServicePortPurpose
api-gateway4100Public entry point
ingress-service4101Channel message intake
orchestrator-service4102Agent runtime and planner
policy-service4103Authorization decisions
execution-manager4104Tool validation and execution
memory-service4105Context and memory storage
node-manager4106Remote node management
audit-service4107Audit trail
approval-service4108Approval workflow

What happens when you send a message​

  1. Ingress receives the message, normalizes it, verifies the source
  2. Orchestrator resolves your identity, retrieves session context, asks the model what to do
  3. Policy evaluates whether any proposed action is allowed
  4. Approval service handles actions that require human sign-off
  5. Execution manager validates the signed intent and runs the action in a sandbox
  6. Memory stores the outcome for future context
  7. Audit records everything that happened

Continue to First Workflow for an end-to-end walkthrough.