Capabilities

Built for agents that actually do things.

Everything below ships in the repository today — verified against the source, not a roadmap.

01

Persistent Memory

Search and reuse relevant context across sessions. FTS5-backed session search with LLM summarization for cross-session recall.

02

Evolving Skills

Skills can be created and improved from experience. The agent writes, refines, and reuses them during real work.

03

Any Model

Use supported providers and custom endpoints without locking the agent to one model. Switch with synapse model.

04

Real Tools

Terminal, filesystem, browser research, integrations, and more — configurable through the built-in tool gateway.

05

Scheduled Work

Run recurring workflows through the built-in cron scheduler, with delivery to any connected platform.

06

Parallel Agents

Delegate independent work to isolated subagents that run in parallel workstreams.

07

Multi-Platform

Interact through the CLI or messaging platforms — Telegram, Discord, Slack, WhatsApp, Signal, Email — via one gateway process.

08

Runs Anywhere

Local machines, containers, remote environments, and supported serverless backends including Modal, Daytona, and Vercel Sandbox.

The learning loop

Most agents forget. Synapse compounds.

Synapse is built around a closed loop: every task produces experience, experience is curated into memory and skills, and those skills improve future execution.

continuous loop
  1. Experience
  2. Memory
  3. Skills
  4. Better execution
  5. More experience
  6. loops back — knowledge persists across sessions

The learning loop is not a side feature — it is the core identity of Synapse. Agent-curated memory, autonomous skill creation after complex tasks, and periodic self-nudges to persist knowledge are all part of the default runtime.