SynapsCLI
A terminal-native AI agent runtime in Rust that runs parallel subagents on many models from one binary.
At a Glance
SynapsCLI core agent runtime released under the Apache License 2.0; free to use, modify and distribute. Model provider costs (subscriptions, API keys, cloud) are external.
Engagement
Available On
Alternatives
Listed Nov 2026
About SynapsCLI
SynapsCLI is an open-source, Apache-2.0 licensed agent runtime built in Rust by HaseebKhalid1507. It runs a crew of AI agents in the terminal through a TUI, on a range of model providers, from a single binary. The latest release listed is v0.11.1.
What It Is
Synaps is a terminal-first agent runtime for interactive chat, parallel agent orchestration, and autonomous supervision. A background daemon owns sessions, so they outlive the terminal window, and several clients can attach to one session to type, steer, or take over. Named subagents work in parallel, and any one can be opened in its own view and steered mid-flight.
Models and Credentials
Models are addressed as provider/model. The README lists Claude, Codex, Grok, Copilot, Gemini, Kimi, Meta Muse, Azure OpenAI, Amazon Bedrock, Google Vertex AI, any OpenAI-compatible endpoint, and local models via Ollama, LM Studio, vLLM or llama.cpp. Sign-in can use a subscription, an API key, cloud identity, or a local endpoint. A credential broker hands agents short-lived, scoped tokens rather than raw keys and supports several accounts per provider. Cloud routes are text-only for now.
Interfaces and Extensibility
Besides the TUI, the CLI offers a line client (attach), a script-friendly send command, a stdin/stdout chat mode for pipes and CI, JSON-RPC over stdio for embedding, and a WebSocket server with token auth. Extensions are folders under ~/.synaps-cli/plugins/ that talk JSON-RPC 2.0 over stdio in any language. Each runs as its own crash-isolated process under a permissions manifest, with lifecycle hooks. MCP servers are supported, and 21 built-in themes are included.
Session Cost and Memory
The README describes a long-lived prompt cache on a Claude subscription, large tool output kept on disk behind a short preview, deduplicated re-reads of unchanged files, and context rollover for workers. It also covers project memory and /compact checkpoints, plus optional caps on rounds, tool calls, time, tokens and cost. The README publishes its own benchmark figures for memory use, measured on one machine.
Tradeoffs to Know
The README states Synaps is terminal-first, not an IDE plugin or desktop app. It is not a sandbox: tools run as the user, so untrusted work should use a container or VM. The project describes itself as young and fast-moving, and recommends reading upgrade notes before upgrading.
Community Discussions
Be the first to start a conversation about SynapsCLI
Share your experience with SynapsCLI, ask questions, or help others learn from your insights.
Pricing
Open Source
SynapsCLI core agent runtime released under the Apache License 2.0; free to use, modify and distribute. Model provider costs (subscriptions, API keys, cloud) are external.
- Apache License 2.0
- Single ~32 MB binary
- Run parallel AI agents in the terminal on any model
- Works with provider subscriptions, API keys, cloud identity or local models (e.g. Ollama)
- Linux, macOS and Windows 10/11 (x64)
Capabilities
Key Features
- Background daemon with persistent, pinnable sessions
- Multiple clients attached to one live session
- Parallel named subagents that can be steered mid-flight
- Multi-provider model support including local models
- Credential broker with short-lived scoped tokens
- Optional caps on rounds, tool calls, time, tokens and cost
- Prompt caching and context rollover for long sessions
- Image, PDF and text attachments
- Extensions in any language over JSON-RPC 2.0
- MCP server support
- JSON-RPC and WebSocket server APIs
- Chat mode for pipes, scripts and CI
- 21 built-in themes
