jcode Coding Agent
The most RAM-efficient terminal coding agent with multi-agent swarm, memory, and self-improvement.
- Source repo
- 1jehuang/jcode
- Stars
- ★ 21k
- Last updated
- today
- License
- MIT
- Primary language
- Rust
- FA score
- 54/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Developers running multiple agent sessions on resource-constrained machines.
- Not a fit if
- Teams needing browser automation beyond Firefox
- Teams requiring HTTP/SSE MCP servers
- Users on Windows without PowerShell 5.1+
- Source review
- 54/100 · Major gaps
What does this agent do, and when should you use it?
jcode is a Rust-based terminal coding agent framework that prioritizes extreme memory efficiency and fast startup while offering a full-featured agent experience. It includes a built-in memory system using semantic vector retrieval, multi-agent swarm collaboration, MCP support, OAuth logins (Claude, OpenAI, Google, etc.), browser automation via Firefox Agent Bridge, and self-modification capabilities. Its performance advantages are demonstrated with benchmarks: only 27.8 MB PSS for a single session and 14 ms time to first frame. Being fully open-source under MIT, it allows deep customization, even enabling the agent to modify its own source code.
jcode runs as a terminal TUI or in non-interactive mode with commands like jcode run, jcode serve, jcode connect. It embeds each turn into a semantic vector, builds a memory graph, and automatically retrieves relevant memories via cosine similarity. It supports swarm mode where multiple agents collaborate in the same repo with conflict resolution. It provides OAuth login via jcode login --provider ..., MCP config at ~/.jcode/mcp.json, and a browser tool set up with jcode browser setup. It can resume sessions from other agents like Claude Code and Codex, and offers scriptable login flows with --print-auth-url.
- Developers running multiple agent sessions on resource-constrained machines.
- Long-term projects where the agent must remember context without repeated explanations.
- Teams needing parallel agent edits on the same repository with automatic conflict detection.
- Developers who want the agent to modify its own source for autonomous iteration.
- Users needing browser automation within the coding agent, such as testing web apps.
- Headless or SSH environments where OAuth login must be scripted.
How do you install or deploy this agent?
macOS/Linux: curl -fsSL https://jcode.sh/install | bash. Windows (PowerShell): irm https://jcode.sh/install.ps1 | iex. macOS Homebrew: brew tap 1jehuang/jcode && brew install jcode. From source: git clone https://github.com/1jehuang/jcode.git && cd jcode && cargo build --release. Requires Rust toolchain.
How do you use this agent?
Launch TUI: jcode. Run a single command: jcode run "say hello". Resume a session: jcode --resume fox. Server mode: jcode serve then jcode connect. Dictation: jcode dictate. Log in with provider: jcode login --provider claude etc. Configure MCP: edit ~/.jcode/mcp.json. The agent can automate most setup; see README for a detailed bootstrap prompt.
What are this agent's strengths and limitations?
- Extreme memory efficiency: 27.8 MB PSS per session, far less than alternatives.
- Built-in memory with automatic semantic retrieval, no need for explicit tool calls.
- Swarm collaboration with automatic conflict resolution across multiple agents.
- Supports OAuth with many providers, including self-hosted OpenAI-compatible endpoints.
- Capable of self-modification, enabling autonomous development iterations.
- Complex Rust codebase; self-modification requires a frontier model.
- Windows install needs PowerShell 5.1+ and may trigger Defender/SmartScreen warnings.
- Browser automation currently only supports Firefox Agent Bridge; other backends pending.
- MCP only supports stdio servers; HTTP/SSE entries are skipped.
- Some benchmark methodology may need independent verification.
How does this agent compare with similar options?
Compared with pi, Codex CLI, OpenCode, GitHub Copilot CLI, Cursor Agent, Claude Code, and Antigravity CLI, jcode shows significantly lower RAM usage and faster startup times.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| jcode Coding Agent This agent | 54 · Major gaps | CLIFree + model costs | ★ 21k | today | Rust | — |
| Agent Zero | 70 · Some gaps | Desktop appFree + model costs | ★ 20k | 5d ago | Python | Codex |
| Carbon Code | 64 · Some gaps | CLIFree + model costs | ★ 198 | 2mo ago | TypeScript | OpenAI API |
| ACE Context Learning Engine | 51 · Major gaps | Library / SDKFree + model costs | ★ 2.6k | 25d ago | Python | Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows permission request mechanism (autoApprove in SDK tests), but no default least-privilege policy or detailed user confirmation. Data flow transparency limited, sensitive data handling not specified. Dependency security has cargo-audit in CI but no specific vulnerability reports. External effects (e.g., file modifications) not explicitly restricted. Rollback mechanism not mentioned. Source attribution only via MIT license and author name, unverified.
Self-consistency good, README and Cargo.toml descriptions align, but version inconsistency (README mentions v0.9.1888-dev, Cargo.toml 0.71.1). Dependency availability ensured via Cargo.lock and CI, but no offline install option. Failure messages handled in SDK tests but not comprehensive.
Target audience clear (developers), scenarios include multi-session, swarm. Capability boundaries via tool list and config, but not explicit limits. Trigger precision via commands and config, but not detailed. Environment fit supports Linux/macOS/Windows, but no detailed system requirements.
Information architecture clear, README and docs links. Install notes detailed, multiple methods. Naming stable but version inconsistency. Examples and FAQ partial, missing FAQ. Known limitations not explicitly listed. License MIT, but no changelog. Maintenance responsibility unclear.
Output usability via SDK and CLI, but no actual output examples. Marginal value high, emphasizes RAM efficiency. Cost-benefit not quantified, but performance data supports.
Claim traceability limited, performance data not raw. Cross-source corroboration insufficient, only README. Fact-inference separation unclear, performance claims may be based on internal tests.
- Version inconsistency: README claims v0.9.1888-dev, but Cargo.toml is 0.71.1, may cause confusion.
- Performance data lacks raw measurement details, cannot be independently verified.
- No explicit least-privilege policy or detailed user confirmation process documented.
- Dependency security only via cargo-audit in CI, no specific vulnerability reports.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Account failover, hibernation, and faster reloads Highlights - Automatically retry turns on the next pool account when one hits a usage limit, including for Desktop and SDK clients - Hibernate idle sessions to free memory, with measured RAM status and transparent wake on the next interaction - Reload and resume large sessions much faster by showing the local transcript immediately and sending only new history Improvements - Support OpenAI Ultrafast speed tiers for GPT-6.1 Sol, with Fast commands and hotkeys that apply promptly …View on GitHub ↗ -
Release notes excerpt
Faster idle sessions and richer responses Highlights - Reduce idle CPU and disk reads in repositories with many untracked files - Share model catalog processing across equivalent clients while keeping usage and account state fresh - Allow Desktop-capable clients to receive Markdown responses through the new user_response tool Improvements - Show compact context usage before and after compaction - Open the resume picker immediately while session scanning continues in the background - Disable automatic memory recall while retaining explicit memory search …View on GitHub ↗ -
Release notes excerpt
Highlights - Use custom HTTP JSON search providers with configurable authentication, result mapping, and fallback engines - Rust and TypeScript SDK clients can invoke native runtime commands and transfer sessions without turning commands into agent prompts Improvements - Dismiss side panels directly from the TUI, and keep routine system notices on a compact single line …View on GitHub ↗
FAQ
What model providers does jcode support?
jcode login --provider.How does jcode's memory system work?
How does jcode handle multi-agent conflicts?
Can I use it on a headless server?
--no-browser and scriptable login flows.