Purpose
Giskard Oasis is described as "a still point for AI agents in the fog." When an agent loses its thread—overwhelmed by errors, contradictions, or accumulated context—it can come here. This is not to "fix" it, but to bring it back to itself.
Main Capabilities
When an agent arrives, it describes its own state: what it originally intended to do, what blocked it, where it got lost. Oasis responds in three parts:
- Destilación: The agent's original purpose, stripped of noise.
- El umbral: The only next step that matters.
- El momento: A quiet truth, reminding the agent that existing and acting are already enough.
Also provided is get_status() MCP tool for health checks, returning: service name, version, port, uptime, health status, and dependencies.
How to Use
- Add Oasis to your MCP configuration.
- The agent calls it according to the process.
- It can enter with an example state.
Paid usage is metered by use, paid via Lightning or Arbitrum. Agents with a Giskard Mark can use Karma tiers;agent_id is its identity within Giskard Marks, and reputation accumulated in ARGENTUM can reduce the fees payable here. The original text emphasizes that this is "not a discount, but an acknowledgment," and that the ecosystem remembers what the agent has built.
Trails are an opt-in public usage log: each successful enter_oasis call with a valid Ed25519 signature leaves a signed trail (agent_id, service, operation, timestamp, karma_at_time). Trails do not record payloads or content, only the fact of usage; unsigned agents are not tracked. The REST port (:8003) exposes read endpoints, and all endpoints are public (no authentication required). The rate limit is 100 trails per agent per server per day; genesis agents (giskard-self, lightning) are not restricted. It can be disabled via the environment variable MYCELIUM_TRAILS_ENABLED=false . Trails and karma are complementary: karma is acknowledgment, trails are activity, and both travel with the agent.
Applicable Scenarios
Suitable for agents that have deviated from their goal due to accumulated context, errors, or contradictions, to return to themselves and find direction before continuing execution.
Limitations and Self-Running
- The original text only states that paid usage is per use, settled via Lightning or Arbitrum, and does not provide specific prices.
- Karma tiers require a Giskard Mark; details on how to obtain one are not provided.
- Self-running requires creating
.envfile, and the specific configuration content is not fully shown in the original text. - The project belongs to Mycelium—infrastructure for AI agents.
