Your team’s knowledge, available to your AI agents.

Firekeep is a shared, self-hosted knowledge and continuity layer for AI coding teams. It combines persistent memory, a Corpus knowledge base, reusable Skills, durable session state, local code intelligence, agent coordination, guided decisions and inspectable traces behind the coding tools your team already uses.

Corpus stores ingested documents and bounded public-documentation crawls as searchable sources that can surface through ordinary Firekeep recall. With a generation model configured, the Knowledge pipeline can also draft procedures for human review; drafts do not become active Skills until they are approved.

When an agent needs several choices from you, the Decision Board opens one local browser review with the agent's questions, relevant memory evidence and any attached visual explanations. Your answers return together to the waiting agent; selecting an option does not execute an action or automatically save it as memory.

Yes. Relay provides tasks, catch-up channels, bulletins, pollable direct messages, best-effort presence and expiring resource leases across your agent identities. Notification timing and hard lease enforcement depend on the lifecycle hooks each coding client exposes.

Symdex is Firekeep's built-in local code-intelligence server. It indexes symbols and relationships across Python, JavaScript, TypeScript, Go, Rust, Java, PHP, C, C#, Ruby, Kotlin and Swift so agents can request focused code, caller and impact context. Firekeep installs it automatically with Solo.

Across 360 direct known-symbol lookups drawn from 285 production source files in 12 pinned repositories, the stable Symdex response had a 54.8% lower estimated token count on average than sending the entire containing file. Retrieved source matched the pinned Git source in 360/360 checks. Results varied: 58 lookups used more tokens, generally where the whole file was already very small. This is a targeted-retrieval benchmark, not a promise about whole-task use. In a separate mixed-task Click benchmark, Symdex used 23% fewer total context tokens, with a judged-accuracy point estimate of 4.42 versus 4.45; that result does not establish statistical equivalence.

The index runs on your workstation, not on the Firekeep server. As with any local MCP tool, the code context an agent requests becomes part of that agent's working context and may be sent to the model provider you configured.

The installer includes adapters for Claude Code, Codex, Kiro and OpenCode. Other MCP-capable clients can be configured manually, but lifecycle automation differs because each client exposes different hook capabilities.

Solo covers one natural person, one workspace and one deployment. That person can use unlimited personal devices, terminals, agent identities and background workers, with the complete current capability set.

No hosted Firekeep service is required. The server, datastores and default inference path run on your infrastructure. Public downloads and update checks use the release channel, and optional connectors or model providers make external requests only when you configure them.

No. A default server binds to loopback. You can reach it with SSH forwarding, a private network such as Tailscale or WireGuard, or your own HTTPS reverse proxy. Firekeep does not require a network vendor.

Team is being prepared now. It will use the same Firekeep product and add multiple people through signed seat entitlements. Join the waitlist to help shape pricing and onboarding.